liangzid/robench2025a_test_all_category_setq-bioSCP-s-50

shuffled_text stringlengths 284 1.5k	A stringclasses 6 values	B stringclasses 6 values	C stringclasses 6 values	D stringclasses 6 values	label stringclasses 4 values
A: Given RhoFold+’s high accuracy and speed, we finally conducted ablation studies to understand which components and information are important to RhoFold+’s predictions. The architectural components we investigated included 4 different modules (Fig.5a, see Methods)B: Ablation studies were performed on 138 PDB targets (collected between April 2022 and December 2023) with sequence similarities below 80% to our training set and lengths ranging from 16 to 300 nt (the ‘Ablation set’). By removing each RhoFold+ component, we observed that all contributed to improving the performance, with the MSA module being the most critical, followed by the RNA-FM language model (Fig.5a)C: The RNA-modified version of AlphaFold2, without the MSA module, performed worse than RhoFold+ (Fig.5a). Notably, removing RNA-FM led to a sharper performance decline for dissimilar sequences (Fig.5b), and the RNA-FM module seemed to compensate for the loss of the MSA module, maintaining higher TM-scores (Fig.5c). Additionally, removing the recycling module most significantly affected predictions for longer sequences, likely due to its role in effectively deepening the model (Supplementary Figure 7; detailed discussion in Supplementary).	BAC	ABC	ACB	BAC	Selection 2
A: CF identified and developed the connection to ecological orbits and simple harmonic motion via the maternal effect. LP drafted the initial manuscript; both SL and LP edited the first version posted on arXiv [24]. CF, SL, and LP edited the final version and code [12]. The authors are ordered alphabetically. B: SL studied the virial theorem while participating in the “Introduction to Astrophysics” cluster in the COSMOS summer program held at UC Irvine from July 9, 2023 to August 4, 2023. Specifically, she used the virial theorem to repeat Zwicky’s Coma cluster mass estimates using modern measurements of velocity dispersion and galaxy positionsC: LP learned of the virial theorem from SL and, in discussing its proof with SL, realized that it must be related to the Price equation. SL and LP explored the applications and implications of the connection	ACB	CAB	BAC	BAC	Selection 2
A: In Tsamardinos et al. six AutoML tools were compared [14]. Of these, only one had a predictive performance estimation strategy that could adjusts for multiple model validations (limitedly to SO problems and not affecting model selection), while most of the tools have the necessity to withold a test set for an unbiased estimation of the performance of the winning model, thus loosing samples from the final model training.B: Although various methods have been developed to enhance performance estimation in model selection using k-fold CV, their design and implementation have been limited to SO problems. Tsamardinos et al. [12] compared double CV, the Tibshirani and Tibshirani method [13], and nested CV in their ability to improve the estimation of the fitness for SO problemsC: These algorithms modify fitness estimations but do not change the model selection process; the chosen model remains the same as it would be using simple hyperparameter optimization with k-fold CV for model evaluation. Automated ML (AutoML) tools offer an approach designed to explore various model and hyperparameter combinations. These tools aim to identify and deliver the most effective model along with an assessment of its performance	ACB	ACB	ACB	CAB	Selection 4
A: In the past few years, state-of-the-art, DNN-based methods were introduced for IVIM parameter estimation. Bertleff et al. [8] demonstrated the ability of supervised DNN to predict the IVIM model parameters from low SNR DWI dataB: Barbieri et al. [6] proposed an unsupervised, physics-informed DNN (IVIM-NET) with results comparable to Bayesian methods with further optimizations by Kaandorp et al. [24] (IVIM-NEToptimoptim{}_{\textrm{optim}}start_FLOATSUBSCRIPT optim end_FLOATSUBSCRIPT)C: Zhang et al. [42] used a multi-layer perceptron with an amortized Gaussian posterior to estimate the IVIM model parameters from fetal lung DWI data. Recently, Vasylechko et al. [37] used unsupervised convolutional neural networks (CNN) to improve the reliability of IVIM parameter estimates by leveraging spatial correlations in the data.	ABC	CAB	BAC	ACB	Selection 1
A: The tasks above concern prediction of static properties, including structuresB: Because molecular dynamics trajectories consist of sequences of structures, GNNs should be useful for identifying features for computing reaction statistics, and several groups have combined GNNs with VAMP [8] to learn metastable states and relaxation time scales of both materials and biomolecules [31, 32, 33, 34, 35]C: These groups report improved variational scores, convergence for shorter lag times, and more interpretable learned representations relative to VAMPnets based on fully connected networks. However, existing GNNs for analyzing dynamics do not readily scale to large numbers of atoms, so the graphs in these studies are small, either because the molecules are small, or only a subset of atoms (e.g., the Cα atoms of proteins) are used as inputs.	ABC	ACB	CBA	ACB	Selection 1
A: The concomitant impact of public health interventions and disease transmission rates will decrease the population’s hospitalization rate, although it increases with the severity of the disease. Our research highlights that quarantine may be a favourable strategy at the individual level in specific epidemiological situations. This may provide insights for public health officials in developing intervention plans to control an epidemic. B: We have shown that a rise in the rate of disease transmission leads to an increase in the illness burden and the load placed on hospitals. Consequently, more individuals will be involved in the decision-making processC: We investigated this decision-making approach by developing a standard SEIR (susceptible-exposed-infected-recovered) model, including hospital and quarantine compartments. We use a fractional derivative approach to represent disease propagation within a population	BCA	BAC	BCA	CBA	Selection 4
A: The estimation of the parameter vector θ𝜃\thetaitalic_θ, as defined in (13), involves maximizing the product of the two likelihood functions (21) and (22)B: To enhance numerical stability, it is often advantageous to consider the formulation in terms of the logarithmic transformation (II-B), which involves maximizing the sum of the logarithms of the two likelihood functions insteadC: This approach seamlessly integrates into the Bayesian estimation framework, allowing for more comprehensive propagation of uncertainty and the incorporation of external information into the statistical model. Consequently, this method yields parameter estimates that fully leverage all available information while accounting for uncertainty in a rigorous manner.	ACB	ACB	BAC	ABC	Selection 4
A: Most notably, we do not include the PVC and DVA neurons in our core set despite their strong association with locomotion and strong connections with the core setB: This is because the PVC and DVA neurons do not appear in the whole-brain imaging datasets we use to fit our model [1], making it infeasible to fit model parameters for these neurons using the methods we employ. Another omission of note is that we exclude motor neurons, many of whichC: We remark that though guided by data, there is some arbitrariness in the choices above as well as practical constraints	ABC	BCA	ACB	BAC	Selection 2
A: Aside from instantaneous energetics, gait transition from walking to running has been attributed to muscle force-velocity behavior [31], interlimb coordination variability [32], mechanical load or stress [33, 18], and cognitive or perceptual factors [34, 35]; see [36] for a reviewB: And importantly, we know it is unclear how any of these ‘instantaneous’ theories with the gradual walk-run-mixture-based gait transition regime behavior we have demonstrated in this manuscript. Predicting a walk-run mixture over a full bout as being optimal must necessarily require a theory that integrates some performance measure over the entire bout. C: However, none of these factors can show why there may be hystereses between the walk-to-run and the run-to-walk speeds on treadmills [37]	BCA	ACB	ABC	CBA	Selection 2
A: This is a different cost from our model, which they supported by showing an approximate quadratic scaling of metabolic cost with force frequencyB: Van der Zee and Kuo [28] proposed a model of metabolic rate proportional to the second derivative of force, which is equivalent to the metabolic cost per movement being proportional to the first derivative of forceC: Our model is roughly consistent with their data, also indicating a roughly quadratic with oscillation frequency, though more specifically our model predicts a faster than quadratic scaling of metabolic cost with frequency when the force mean and amplitude are fixed (γ2>2subscript𝛾22{\gamma_{2}>2}italic_γ start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT > 2). Reviewing Van der Zee and Kuo’s data (figure [28]) suggests their data may also be consistent with a slightly faster-than-quadratic scaling with oscillation frequency. In future work, we will consider how well alternative models with higher derivatives fit our or even more diverse data.	BAC	CBA	BCA	ACB	Selection 1
A: Common simplifications like considering a population-wide “contact rate” do not take this structure into account. Considering this structure leads to more realistic models of the epidemic. B: This trend also intensified during the Covid-19 pandemic, where network based epidemic models were applied and developed, in an attempt to prevent the spread of the disease [17, 18]. While classical deterministic models assume a homogeneously-mixed population, in practice populations are not fully-mixed and more likely to behave as a complex network [19]C: Namely, in a real population diseases spread between individuals only when they are actually in contact with one-another. Therefore, when modeling epidemics one should consider the population structure	CAB	BAC	ABC	BAC	Selection 1
A: This binding can be represented as p⁢H⁢L⁢Ah⁢a⁢p⁢l⁢o𝑝𝐻𝐿subscript𝐴ℎ𝑎𝑝𝑙𝑜pHLA_{haplo}italic_p italic_H italic_L italic_A start_POSTSUBSCRIPT italic_h italic_a italic_p italic_l italic_o end_POSTSUBSCRIPT rather than p⁢H⁢L⁢Aa⁢l⁢l⁢e⁢l⁢e𝑝𝐻𝐿subscript𝐴𝑎𝑙𝑙𝑒𝑙𝑒pHLA_{allele}italic_p italic_H italic_L italic_A start_POSTSUBSCRIPT italic_a italic_l italic_l italic_e italic_l italic_e end_POSTSUBSCRIPTB: the haplotype as a wholeC: As far as we know, HLA molecules are	ACB	ABC	BAC	CAB	Selection 3
A: In summary, we have shown that fast antigenic drift can induce selection for complex adaptive immune responses, in line with our general picture of complexity in dynamical recognition systemsB: These processes depend on the long-term co-evolution of host immune systems and multiple viral pathogens [86], which is beyond the scope of this paper. Other features of immune recognition not contained in the minimal model include a feedback of recognition onto the speed of target evolution. If epitope mutations come with a collateral cost to the pathogen (for example, through their effect on protein stability), a complex immune response can slow down or halt antigenic drift; this effect has recently been reported for measles [76].C: This finding is consistent with observations: several fast-evolving RNA viruses, including influenza [25, 82], norovirus [69], and Sars-Cov-2 [83, 84, 85], are subject to multi-epitope responses, although successful targeting of a single epitope is sufficient for neutralization [75]. However, the minimal model is insufficient to describe the full dynamics of immune recognition complexity	CAB	BAC	ACB	ABC	Selection 3
A: Table 3: Relative expression (RE) of genes in TLL1-transduced MT-2 cells over MT-2 cells transduced with an empty vectorB: Genes were picked up if the RE of type 1 was larger than 11111111 and the RE of type 2 was smaller than 3333 or if the RE of type 1 was smaller than 1111 and the RE of type 2 was larger than 5555.C: Type 1(2) corresponds to MT-2 cells transduced with TLL1-isoform 1 (isoform 2)	BCA	ACB	CAB	CAB	Selection 2
A: (2019). Our study expands the concept of brain metastable states to multi-person interactions, providing a novel perspective without relying on millisecond time-scale synchronizationB: Our work offers a fresh perspective on inter-brain connectivity by analyzing a hyperscanning neuroimaging dataset through multi-brain symbolic dynamics. Metastable states dynamics are crucial for brain flexibility, enabling seamless transitions between stable functional states and supporting cognitive processes and resilience to noise through robust mechanismsRabinovich, Zaks, and Varona (2020); Tognoli and Kelso (2014); Kelso and Tognoli (2006); Roberts et alC: Our results and analytical approach not only broaden the understanding of brain metastable states in the context of human interaction and coordination but also open new avenues for studying the brain mechanisms underlying these processes. This new tool bridges brain dynamics and behavior, offering a powerful and flexible method to explore the intricate neural choreography of multi-person interactions in a broad range of cognitive functions and collaborative behaviors.	BCA	BAC	ABC	CBA	Selection 2
A: The output special nodes and pair embedding of Co-Former are employed dependent on different tasks, including two pre-training tasks and two downstream affinity tasks. CN, PN, and RN are the special nodes for complex, protein, and RNA, respectively.B: Figure 2: Overview of CoPRA. Given a protein-RNA complex as input, the sequence information of protein and RNA are fed into a PLM and an RLM, respectivelyC: The output embeddings are selective with interface information and are fed into Co-Former with pairwise information. The Co-Former fuses the 1D and pair embedding by structure-guided multi-head attention and outer product modules, with a task-dependent attention mask	ACB	ABC	CAB	ACB	Selection 3
A: Artifacts in this dataset include out-of-focus areas, tissue folds, cover slips, air bubbles, pen marks, and areas with extreme over- or under-staining, which were carefully labeled to aid in quality control for pathology image analysis. We trained our network using 1500 artifact-free patches from epithelium and stroma classes. We then use this network to classify 474 artifact patches from 474 clean patches.B: HistoROI dataset is developed to segment WSIs into six key classes: epithelium, stroma, lymphocytes, adipose, artifacts, and miscellaneousC: For the experiment to filter out low-quality noisy image patches from good quality ones, we used HistoROI dataset [11]	ACB	CBA	ABC	BCA	Selection 2
A: Empirical validation of the urban scaling hypothesis has driven numerous studies using data from multiple countries and examining a wide array of urban indicators. In the realm of disease incidence, the seminal work by Bettencourt et al. reported that HIV/AIDS cases in the USA scale superlinearly with population size [9], a pattern also observed in Brazil by Antonio et al. [12]B: The authors suggest that the superlinear scaling observed in some infectious diseases likely reflects the increased number of contacts among residents in larger cities compared to smaller towns [14]. In contrast, the isometric and sublinear scaling of other infectious diseases may be due to the insufficient allocation of medical resources in smaller cities. Patterson-Lomba et al. [15] reported that sexually transmitted diseases scale superlinearly with the population of urban areas in the USA, even when controlling for socioeconomic variables. They also found that income inequality positively correlates with disease incidence, while educational level is negatively correlated.C: Rocha et al. [13] extended this analysis by investigating the scaling of various health indicators using data from Brazil, Sweden, and the USA. Their findings revealed that infectious diseases such as HIV/AIDS, chlamydia, and influenza generally scale superlinearly with population size. However, other infectious diseases, including leprosy, viral hepatitis, and dengue, exhibited isometric or even sublinear scaling	ABC	BCA	CBA	ACB	Selection 4
A: The results show the promising value of GAN-TAT, both through computational performance and validation against clinical data. However, inherent limitations of PIN embedding, such as imbalance, sparsity, and potential overfitting, could effect GAN-TAT’s efficacy. Despite these challenges, this work provided an innovative solution to this topic and offers a promising direction for continued research and enhancement. B: Utilizing the ImGAGN structure, GAN-TAT helps mitigate challenges associated with the use of PINC: This study introduces GAN-TAT, a novel machine-learning framework that leverages network embedding for identifying druggable genes	ABC	CBA	CAB	ACB	Selection 2
A: In order to capture the hierarchical nature of GO labels, labels are represented as a directed acyclic graph and embedded as an additional feature input. They were able to achieve state-of-the-art results for 2 out of 3 problem subclasses and competitive results with NetGO and GOLabeler for the remainder. Notably, pre-training the Transformer through unsupervised training did not lead to performance improvements. B: [91] uses bidirectional LSTMs to condense input sequences into a fixed-size feature vector, which is then used to classify GO functional labels, although performance is not compared with [59, 58]. TALE uses the Transformer architecture [18] to perform GO label classification but with an additional label embeddingC: More sophisticated architectures such as RNNs and Transformers have also been used. One of the challenges from dealing with protein sequences is the large variation in sequence length	ACB	ACB	BAC	CBA	Selection 4
A: VNN architecture and trainingB: The VNN model was pre-trained on a healthy population to glean information about healthy agingC: To facilitate ΔΔ\Deltaroman_Δ-Age that is transparent and methodologically interpretable, we used a multi-layer VNN model that yielded representations from the input cortical thickness features at the final layer, such that the unweighted mean of these representations formed the estimate for chronological age. Details on how this choice of architecture leads to anatomically interpretable ΔΔ\Deltaroman_Δ-Age are discussed subsequently.	ABC	BAC	BCA	BCA	Selection 1
A: Now, we discuss the conditions regarding the non-existence of spatially heterogeneous steady statesB: For this purpose, first, we deduce a priori estimates for non-negative solutions of the system (17).C: The following two lemmas can be found in [37, 65], respectively	ABC	BAC	CAB	ACB	Selection 4
A: Building on this, Wu et al. (2023, CcBHLA) used BiLSTM, while TransPHLA (Chu et al., 2022; 2021) incorporated self-attention modules to capture complex dependenciesB: Ye et al. (2023, STMHCpan) modeled peptide-MHC interactions as graphs, introducing graph neural networks to the field. C: With advances in natural language processing, sequence modeling techniques have gained popularity. NetMHCpan (Borole & Rajan, 2024; Reynisson et al., 2020; Jurtz et al., 2017) applied LSTM to both MHC and peptide sequences	ACB	BAC	BCA	CBA	Selection 3
A: The data generated and used within this work can be found at the Open Science Framework repository (Lluís Hernández-Navarro, Kenneth Distefano, Uwe CB: 2024. Supplementary data, code, and videos for “Slow spatial migration can help eradicate cooperative antimicrobial resistance in time-varying environments”. OSF. https://doi.org/10.17605/OSF.IO/EPB28).C: Täuber, and Mauro Mobilia	ACB	BCA	CAB	CAB	Selection 1
A: (2021) and how to identify the effect of ionization on receptor proteins. The examples reproduce existing results, confirming the reliability of our approach, but also demonstrate additional new discoveries made possible by systematic comparison of molecular simulations.B: (2020) how to quantify the influence of force-field parameter changes on simulations of nucleic acids,Cruz-León et alC: Here, we discuss the functionality included in PENSA and demonstrate its usefulness on three real-world applications: We show how to describe the influence of local frustration on loop opening during the catalytic cycle of an oxidoreductase,Stelzl et al	CAB	ABC	CAB	CBA	Selection 4
A: Figure 4 represents the scalp topographies of the grand average difference ERPB: We can see from figure 4 that the effect starts being significant from the timepoint 500 ms to 800 ms (p<0.001𝑝0.001p<0.001italic_p < 0.001) at the parietal region. In each topography, there is also another significant cluster at the frontal region which likely represents the negative polarity of the effect. C: The areas marked by black dots indicate significant differences in effects between target and non-target ERPs as determined by the spatial cluster permutation test described in 2.7, using the same threshold selection procedure as mentioned above	ABC	BAC	ACB	CBA	Selection 3
A: An extensive body of theoretical — and some experimental — work on proofreading reports the intuitive speed-accuracy trade-off. How might our work, motivated by experimental observations of stalling [33, 34, 35, 36, 45, 38], be consistent with such prior work? Prior studies consider only the impact of a few mutations or one biophysical perturbation, e.gB: changing Mg2+ levels [46]C: Such tests cannot probe the nature of any potential tradeoff between speed and fidelity. More precisely, trade-offs manifest as a Pareto front defined by an inequality between traits such as speed and accuracy.	CBA	BCA	ABC	CBA	Selection 3
A: DICOM pixel intensities were rescaled, and the image was resized to a smaller size of 224 x 224 to fit on the GPU. A pretrained Efficient-Net model was used for image processing, and extracted features were then concatenated with LSTM/Neural CDE’s features and passed through a final fully connected linear layer to give the final output.B: Similar preprocessing steps were performed on the structured data but for the multimodal data, each training step involved the incorporation of a random slice of the subject’s baseline CT scanC: To test the hypothesis for the improvement in both LSTM and Neural-CDE performance, a baseline CT scan image was used	ACB	ACB	CBA	CAB	Selection 3
A: Figure S11: The FI predicts future FPFP5 deficits better than NFPFP5. HRS, sex-stratifiedB: The curves are similar between sexes. Visually, individual points appear to agree between sexes within errorC: The AUC is the probability that a metric will correctly rank positive individuals as higher than negative individuals [31] (0.5 is guess, dotted line; 1 is oracle). Note: the x-axes are sorted by AUC. Leave-one-out excludes the outcome deficit from the predictor (doesn’t affect FP frailty prediction). Error bars are standard errors.	BCA	BCA	ABC	CAB	Selection 3
A: (2023), which includes genomic sequences from known organisms—both from human genomes and a curated selection of genomes from multiple species (e.g., fungi, mammalian, invertebrate, bacteria)—and shuffle it into our metagenomic reads at a 1:8 ratio. B: For this, we sample sequences from the dataset provided by Zhou et alC: The modified training distribution aims to allow for us to maintain performance on metagenomic tasks, such as metagenomic embedding and classification, while also achieving improved performance on a broader set of genomic tasks (i.e., tasks involving non-metagenomic data)	CAB	CBA	BAC	BCA	Selection 2
A: Specifically, DNA methylation levels change rapidly during early development and adolescence, stabilize during adulthood, and may alter again in older ages [3, 4]. This nonlinearity[5] poses significant challenges for traditional age prediction models, which often assume a uniform rate of change across the lifespanB: However, modeling DNA methylation age prediction effectively is complex due to two intrinsic properties of CpG methylation. Firstly, the correlation between CpG site methylation levels and age is not static but changes with age, which we call Epigenetic Correlation Drift (ECD)C: Consequently, relying on a static overall linear correlation value can overlook CpG sites that are highly correlated with age in specific ranges, such as 40-50 years, but exhibit low correlation outside these ranges. Different CpG sites exhibit distinct methylation change patterns, called Heterogeneity Among CpGs (HAC). For example, a CpG site might have a high correlation with age during one age stage but a low correlation in the next, while other CpG sites maintain a high correlation in the subsequent stage.	BAC	BCA	CAB	BCA	Selection 1
A: Furthermore, we compute the reconstruction error as a measure of encoding quality (for details see Appendix E1) and observe that the continuous model is more unstable in terms of representation quality (FigB: This may be due to a continuous drift of neuronal selectivity, which does not appear to be a feature of the discrete model (see Appendix figures F.7 and F.8). We conclude that a sufficiently long hold period and waiting for the stable state of the dynamics is important as recurrent dynamics remove the statistical dependencies and redundancies in the neural activity. This, in turn, allows learning factorised representations.C: 1D)	BCA	BCA	BAC	ACB	Selection 4
A: We demonstrate the wide applicability of UniGuide by tackling a variety of geometry-constrained drug discovery tasksB: With performance either on par with or superior to tailored models, we conclude that UniGuide offers advantages beyond its unificationC: Firstly, while the novelty of conditional models often stems from the condition incorporation, our method redirects focus to advancing unconditional generation, which directly benefits multiple applications. Furthermore, this separation of model training and conditioning allows us to tackle tasks with minimal data, a common scenario in the biological domain.	ACB	BAC	ABC	BCA	Selection 3
A: DynamicBind predicts the ligand-specific protein-ligand complex structure with a deep equivariant generative model.B: NeuralPlexer (Qiao et al., 2023) incorporates essential biophysical constraints and a multi-scale geometric deep learning system for the diffusion processC: For generating the ligand-specific protein-ligand complex structure, a deep equivariant generative model named DynamicBind (Lu et al., 2024) is employed	CAB	CBA	ACB	ACB	Selection 1
A: SCN9A encodes the Nav1.7 sodium channel, which plays a crucial role in pain perception and anesthetic analgesia. CNR1 encodes the cannabinoid receptor type 1, influencing pain and the effects of cannabinoid anesthetics. SLC6A3 and SLC6A5 encode dopamine and glycine transporters, respectively, which are involved in neurotransmission and anesthetic mechanisms.B: Figure 4 presents the ADMET screening results for datasets involving GABRA5, SCN9A, CNR1, SLC6A3, and SLC6A5C: Specifically, GABRA5 encodes the α⁢5𝛼5\alpha 5italic_α 5 subunit of the GABAA receptor, which is targeted by benzodiazepines and barbiturates for sedation	CAB	BCA	BAC	CBA	Selection 1
A: We systematically trained the model to reconstruct images using sets of 4, 16, 32, and 64 learnable attention mapsB: This approach allows us to evaluate the optimal number of features required for effective reconstructionC: The quality of the reconstructions is quantitatively compared with a baseline model, based on the end-to-end reconstruction model of Shen et al. [11]. Additionally, we assess the consistency of the spatial and feature inverse receptive fields (feature IRFs) across different stimuli by computing the standard deviation. Lastly, we employ a data-driven approach to visualize changes in feature RFs, using t-SNE for dimensionality reduction to explore whether electrodes exhibit preferences for certain features.	CBA	BAC	ACB	ABC	Selection 4
A: One great source of explanations is illusions and disorders of body representation. The human brain constantly sends and receives a flow of multisensory informationB: We’ve already finished defining our model and we want to get a further step on rationality. For human beings, the most classical perspective we get self-conscious experience is our bodyC: The integration of this information is responsible not only for the way our body is represented but also for the way it is consciously experienced. Recent studies in cognitive neuroscience have shown that it is possible to modulate bodily self-conscious experience by experimentally changing these multisensory bodily signals. For example, Indian neuropsychologist Vilayanur Ramachandran designed an interesting experiment using a mirrored box to generate synesthesia and kinesthetic illusions in phantom limbs.	BAC	ACB	ABC	ACB	Selection 1
A: demonstrate their emergence within a relatively simple setup: a ring network of N𝑁Nitalic_N FHN units with symmetric connections but non-uniform parametersB: Contrary to the assumption that complex networks are essential for chimera states, Omelchenko et alC: The dynamics of each node in the network are described by:	BAC	ACB	ABC	ACB	Selection 1
A: 6, a simulation within a one-dimensional linear system illustrates the effect of a driving traveling pulse (panel A) - characterized by a constant velocity and width, and a heaviside-like profile - on an oscillatory system with a period of T≈186𝑇186T\approx 186italic_T ≈ 186B: In FigC: The interaction results in phase patterning, as evidenced by a wavelength (λ≈1860𝜆1860\lambda\approx 1860italic_λ ≈ 1860) (panel B). Notably, the regions encountering the driving wave near the critical bottleneck zone exhibit pronounced phase differences, aligning with theoretical expectations (Panel C).	BAC	ABC	CBA	CBA	Selection 1
A: However, its limited reconstruction performance constrains its applicability. Lin et alB: More closely related to our work, Gao et al. (2023) adapts VQ-VAE (van den Oord et al., 2017) for protein structuresC: (2023a) explores discrete structural representation learned by such VQ-VAEs (van den Oord et al., 2017), while Liu et al. (2023) trains a diffusion model on the discrete latent space derived from this approach. Similarly, and concurrently with our work, Liu et al. (2024a) combines finite scalar quantization (Mentzer et al., 2024) with a specialized transformer-based autoencoder for proteins, RNA, and small molecules.	BAC	BCA	ACB	CAB	Selection 1
A: A modified version [8] of the de Gruijl model [7] was implemented in Arbor-Pycat. The morphology of mouse IO neuron C4A was used [25]B: For programming ease, these default values were scaled by the parameter θ𝜃\thetaitalic_θ, which was thus a vector of ones, initially. Optimization was performed against the loss-function:C: After discretization, it contains 4 axonal, 15 dendritic and 2 somatic compartments. Initial parameters are listed in Table 1	ABC	ACB	CBA	CAB	Selection 2
A: Phages infect host cells by adsorbing (attaching) to receptors on the host cell wall and then delivering the genomic content into the host cytoplasm. Phages are much smaller than bacteria and each host cell presents multiple receptors that phages can bind to, so multiple phages can adsorb to a single host cell, though not all adsorptions necessarily lead to infectionB: Multiple adsorptions become increasingly likely at higher phage densities (Turner and Duffy, 2008; Christen et al., 1990) and can become the dominant transmission mode at sufficiently high densities (Turner and Chao, 1999). If phage densities are very high, it is possible that multiple phages simultaneously adsorb to and then infect the same host cellC: Here, we explore the impact of simultaneous infections on phage-host ecology. We define simultaneous infection as infections that occur within a very small time window and distinguish between simultaneous infection and previously studied forms of co-infection, where after a pause an already infected host cell is infected again. Interestingly, given sufficient time phages can prevent multiple, sequential infections through host cell manipulations (Joseph et al., 2009) but these mechanisms are not applicable to the small time window relevant for simultaneous infections.	BAC	BCA	ABC	BAC	Selection 3
A: In this chapter, I have provided a brief introduction to neural networks consisting of DNA, using as an example the winner-take-all network proposed in Ref. Cherry and Qian (2018). The input data is provided as a DNA strand and is processed via biochemical reactionsB: (2013). Such approaches constitute a promising starting point for the development of intelligent matter based on biological materials, and might also find applications in, for instance, medical contexts where input data is already present in a biochemical form.C: On this basis, it is possible to recognize handwritten digits using DNA. Moreover, I have briefly discussed a proposal for DNA-based reservoir computing Goudarzi et al	BCA	BAC	CBA	ACB	Selection 4
A: Positive correlations are observed in the temporal and parietal regions, indicating theta rhythms’ role in auditory information processing and memory integrationB: The Text model shows reduced but still present positive correlations in the temporal lobe, highlighting the involvement of theta rhythms in cognitive processing of text-encoded auditory stimuli. For the Alpha band (8-12 Hz), the Audio model shows positive correlations in the occipital and parietal regions, consistent with alpha rhythms’ association with relaxed states and sensory processing. Negative correlations are prominent in the frontal cortex, suggesting active suppression of irrelevant information during auditory processing. The Text model exhibits a similar pattern with pronounced negative correlations in the frontal regions, indicating alpha rhythms are engaged during both auditory and text-encoded auditory processing, with significant involvement of cognitive control regions.C: The Theta band (4-8 Hz) reveals significant negative correlations in the frontal regions for both models, particularly in the prefrontal cortex, involved in working memory and executive functions during auditory processing	BCA	ABC	ABC	ACB	Selection 1
A: Even when the initial position of the agent is uniformly distributed throughout the system, the agent is not able to learn about its complete environment and gets trapped in nearest attractant peak, for low exploration rate and high learning rateB: When the attractant profile has peaks of different sizes, the exploration-exploitation trade off is even more pronouncedC: As a result, even after a long time has passed, the agent may show same affinity for the lower and higher attractant peaks, which is detrimental for its performance. An optimum balance between exploitation and exploration helps the agent overcome the trapping effect and show its best performance. For non-uniform initial condition, when the agent starts near the lower attractant peak, its probability to localize near the higher peak after a long time, shows a maximum for intermediate rates of exploration and learning.	CAB	CBA	BAC	BCA	Selection 3
A: We plotted the correlation values for all the best performing models from conv1, conv2 and conv3 layers. Each point represents a signal-wise model, and the color represents the model type (blue is AFRT, red is Linear-AlexNet)B: Our analysis shows that the predicted activity from the AFRT model correlate higher with ground truth signals compared to the predicted activity from the baseline model Linear-AlexNet (Fig. 2)C: Although the baseline model makes use of a significant higher amount of features, our results show that models containing the affine components perform better (with a correlation value of 0.5 or higher). Overall, AFRT encodes MUA activity more accurately than the Linear-AlexNet model and our results also show that AFRT is less prone to overfitting.	BAC	ABC	ACB	CBA	Selection 1
A: Genes that have been previously shown to be downregulated in HD from which we obtain a negative correlation between SHAP and gene expression in SPN clusters are: Pde10A [19] or Scnb4 [20]. We have validated other lower rank genes previously described in different HD models, such as Penk1 (cluster iSPN), Gria3 (clusters iSPN, dSPN) [21] or Nrg1 [22]. Although our main focus was put on SPN clusters, other neuronal clusters are worth mentioning since they show shared altered genes with SPN clusters, eg. Gria3 and Nrg1 in clusters eSPN and adult NPC, Onecut2 in eSPN Adam12, Pcp4 in eSPN Rxrg and Adult NPC, or Pde10A in eSPN, Interneuron Whrn, Cholinergic Adrenergic neurons. B: Further model validation was made from SHAP results based on published data obtained from in vivo HD models or patients. For this validation, we compare the correlation between SHAP values and gene expression of previously described altered genes, expecting to observe a positive correlation between higher HD expression levels with positive SHAP values and a negative correlation between reduced HD expression levels with negative SHAP valuesC: Among the top 20 informative genes, some were previously described as being upregulated in HD. For them, we observe a positive correlation between SHAP and gene expression in SPN clusters are: Onecut2 [17] or Sfmtb2 [18]	BAC	CAB	BAC	BCA	Selection 2
A: For instance, the file parsing is highly optimized and executed with asynchronous buffersB: All data is read in streams, so that the number of input files, and their sizes, do not significantly affect the amount of required memory. Processor-intensive steps, such as file parsing and the statistics computations, are multi-threaded with a shared thread pool to leverage modern multi-core systems, and we paid close attention to selecting appropriate data structures for efficiency.C: Performance in runtime and memory was one of the major design goals	CBA	ACB	CAB	BCA	Selection 4
A: Our training data was obtained by extracting 50,000 pairs of conformations and molecules from this datasetB: The GEOM-QM9 dataset is an extension of the original QM9 dataset, enriched with multiple conformations for each molecule, primarily focusing on smaller compounds with a maximum of 9 heavy atomsC: For testing purposes, we reserved 17,813 conformations derived from 150 molecular structures.	CBA	BCA	ACB	BAC	Selection 4

**A**: Given RhoFold+’s high accuracy and speed, we finally conducted ablation studies to understand which components and information are important to RhoFold+’s predictions. The architectural components we investigated included 4 different modules (Fig.5a, see Methods)**B**: Ablation studies were performed on 138 PDB targets (collected between April 2022 and December 2023) with sequence similarities below 80% to our training set and lengths ranging from 16 to 300 nt (the ‘Ablation set’). By removing each RhoFold+ component, we observed that all contributed to improving the performance, with the MSA module being the most critical, followed by the RNA-FM language model (Fig.5a)**C**: The RNA-modified version of AlphaFold2, without the MSA module, performed worse than RhoFold+ (Fig.5a). Notably, removing RNA-FM led to a sharper performance decline for dissimilar sequences (Fig.5b), and the RNA-FM module seemed to compensate for the loss of the MSA module, maintaining higher TM-scores (Fig.5c). Additionally, removing the recycling module most significantly affected predictions for longer sequences, likely due to its role in effectively deepening the model (Supplementary Figure 7; detailed discussion in Supplementary).

BAC

ABC

ACB

BAC

Selection 2

**A**: CF identified and developed the connection to ecological orbits and simple harmonic motion via the maternal effect. LP drafted the initial manuscript; both SL and LP edited the first version posted on arXiv [24]. CF, SL, and LP edited the final version and code [12]. The authors are ordered alphabetically. **B**: SL studied the virial theorem while participating in the “Introduction to Astrophysics” cluster in the COSMOS summer program held at UC Irvine from July 9, 2023 to August 4, 2023. Specifically, she used the virial theorem to repeat Zwicky’s Coma cluster mass estimates using modern measurements of velocity dispersion and galaxy positions**C**: LP learned of the virial theorem from SL and, in discussing its proof with SL, realized that it must be related to the Price equation. SL and LP explored the applications and implications of the connection

ACB

CAB

BAC

Selection 2

**A**: In Tsamardinos et al. six AutoML tools were compared [14]. Of these, only one had a predictive performance estimation strategy that could adjusts for multiple model validations (limitedly to SO problems and not affecting model selection), while most of the tools have the necessity to withold a test set for an unbiased estimation of the performance of the winning model, thus loosing samples from the final model training.**B**: Although various methods have been developed to enhance performance estimation in model selection using k-fold CV, their design and implementation have been limited to SO problems. Tsamardinos et al. [12] compared double CV, the Tibshirani and Tibshirani method [13], and nested CV in their ability to improve the estimation of the fitness for SO problems**C**: These algorithms modify fitness estimations but do not change the model selection process; the chosen model remains the same as it would be using simple hyperparameter optimization with k-fold CV for model evaluation. Automated ML (AutoML) tools offer an approach designed to explore various model and hyperparameter combinations. These tools aim to identify and deliver the most effective model along with an assessment of its performance

ACB

CAB

Selection 4

**A**: In the past few years, state-of-the-art, DNN-based methods were introduced for IVIM parameter estimation. Bertleff et al. [8] demonstrated the ability of supervised DNN to predict the IVIM model parameters from low SNR DWI data**B**: Barbieri et al. [6] proposed an unsupervised, physics-informed DNN (IVIM-NET) with results comparable to Bayesian methods with further optimizations by Kaandorp et al. [24] (IVIM-NEToptimoptim{}_{\textrm{optim}}start_FLOATSUBSCRIPT optim end_FLOATSUBSCRIPT)**C**: Zhang et al. [42] used a multi-layer perceptron with an amortized Gaussian posterior to estimate the IVIM model parameters from fetal lung DWI data. Recently, Vasylechko et al. [37] used unsupervised convolutional neural networks (CNN) to improve the reliability of IVIM parameter estimates by leveraging spatial correlations in the data.

ABC

CAB

BAC

ACB

Selection 1

**A**: The tasks above concern prediction of static properties, including structures**B**: Because molecular dynamics trajectories consist of sequences of structures, GNNs should be useful for identifying features for computing reaction statistics, and several groups have combined GNNs with VAMP [8] to learn metastable states and relaxation time scales of both materials and biomolecules [31, 32, 33, 34, 35]**C**: These groups report improved variational scores, convergence for shorter lag times, and more interpretable learned representations relative to VAMPnets based on fully connected networks. However, existing GNNs for analyzing dynamics do not readily scale to large numbers of atoms, so the graphs in these studies are small, either because the molecules are small, or only a subset of atoms (e.g., the Cα atoms of proteins) are used as inputs.

ABC

ACB

CBA

ACB

Selection 1

**A**: The concomitant impact of public health interventions and disease transmission rates will decrease the population’s hospitalization rate, although it increases with the severity of the disease. Our research highlights that quarantine may be a favourable strategy at the individual level in specific epidemiological situations. This may provide insights for public health officials in developing intervention plans to control an epidemic. **B**: We have shown that a rise in the rate of disease transmission leads to an increase in the illness burden and the load placed on hospitals. Consequently, more individuals will be involved in the decision-making process**C**: We investigated this decision-making approach by developing a standard SEIR (susceptible-exposed-infected-recovered) model, including hospital and quarantine compartments. We use a fractional derivative approach to represent disease propagation within a population

BCA

BAC

BCA

CBA

Selection 4

**A**: The estimation of the parameter vector θ𝜃\thetaitalic_θ, as defined in (13), involves maximizing the product of the two likelihood functions (21) and (22)**B**: To enhance numerical stability, it is often advantageous to consider the formulation in terms of the logarithmic transformation (II-B), which involves maximizing the sum of the logarithms of the two likelihood functions instead**C**: This approach seamlessly integrates into the Bayesian estimation framework, allowing for more comprehensive propagation of uncertainty and the incorporation of external information into the statistical model. Consequently, this method yields parameter estimates that fully leverage all available information while accounting for uncertainty in a rigorous manner.

ACB

BAC

ABC

Selection 4

**A**: Most notably, we do not include the PVC and DVA neurons in our core set despite their strong association with locomotion and strong connections with the core set**B**: This is because the PVC and DVA neurons do not appear in the whole-brain imaging datasets we use to fit our model [1], making it infeasible to fit model parameters for these neurons using the methods we employ. Another omission of note is that we exclude motor neurons, many of which**C**: We remark that though guided by data, there is some arbitrariness in the choices above as well as practical constraints

ABC

BCA

ACB

BAC

Selection 2

**A**: Aside from instantaneous energetics, gait transition from walking to running has been attributed to muscle force-velocity behavior [31], interlimb coordination variability [32], mechanical load or stress [33, 18], and cognitive or perceptual factors [34, 35]; see [36] for a review**B**: And importantly, we know it is unclear how any of these ‘instantaneous’ theories with the gradual walk-run-mixture-based gait transition regime behavior we have demonstrated in this manuscript. Predicting a walk-run mixture over a full bout as being optimal must necessarily require a theory that integrates some performance measure over the entire bout. **C**: However, none of these factors can show why there may be hystereses between the walk-to-run and the run-to-walk speeds on treadmills [37]

BCA

ACB

ABC

CBA

Selection 2

**A**: This is a different cost from our model, which they supported by showing an approximate quadratic scaling of metabolic cost with force frequency**B**: Van der Zee and Kuo [28] proposed a model of metabolic rate proportional to the second derivative of force, which is equivalent to the metabolic cost per movement being proportional to the first derivative of force**C**: Our model is roughly consistent with their data, also indicating a roughly quadratic with oscillation frequency, though more specifically our model predicts a faster than quadratic scaling of metabolic cost with frequency when the force mean and amplitude are fixed (γ2>2subscript𝛾22{\gamma_{2}>2}italic_γ start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT > 2). Reviewing Van der Zee and Kuo’s data (figure [28]) suggests their data may also be consistent with a slightly faster-than-quadratic scaling with oscillation frequency. In future work, we will consider how well alternative models with higher derivatives fit our or even more diverse data.

BAC

CBA

BCA

ACB

Selection 1

**A**: Common simplifications like considering a population-wide “contact rate” do not take this structure into account. Considering this structure leads to more realistic models of the epidemic. **B**: This trend also intensified during the Covid-19 pandemic, where network based epidemic models were applied and developed, in an attempt to prevent the spread of the disease [17, 18]. While classical deterministic models assume a homogeneously-mixed population, in practice populations are not fully-mixed and more likely to behave as a complex network [19]**C**: Namely, in a real population diseases spread between individuals only when they are actually in contact with one-another. Therefore, when modeling epidemics one should consider the population structure

CAB

BAC

ABC

BAC

Selection 1

**A**: This binding can be represented as p⁢H⁢L⁢Ah⁢a⁢p⁢l⁢o𝑝𝐻𝐿subscript𝐴ℎ𝑎𝑝𝑙𝑜pHLA_{haplo}italic_p italic_H italic_L italic_A start_POSTSUBSCRIPT italic_h italic_a italic_p italic_l italic_o end_POSTSUBSCRIPT rather than p⁢H⁢L⁢Aa⁢l⁢l⁢e⁢l⁢e𝑝𝐻𝐿subscript𝐴𝑎𝑙𝑙𝑒𝑙𝑒pHLA_{allele}italic_p italic_H italic_L italic_A start_POSTSUBSCRIPT italic_a italic_l italic_l italic_e italic_l italic_e end_POSTSUBSCRIPT**B**: the haplotype as a whole**C**: As far as we know, HLA molecules are

ACB

ABC

BAC

CAB

Selection 3

**A**: In summary, we have shown that fast antigenic drift can induce selection for complex adaptive immune responses, in line with our general picture of complexity in dynamical recognition systems**B**: These processes depend on the long-term co-evolution of host immune systems and multiple viral pathogens [86], which is beyond the scope of this paper. Other features of immune recognition not contained in the minimal model include a feedback of recognition onto the speed of target evolution. If epitope mutations come with a collateral cost to the pathogen (for example, through their effect on protein stability), a complex immune response can slow down or halt antigenic drift; this effect has recently been reported for measles [76].**C**: This finding is consistent with observations: several fast-evolving RNA viruses, including influenza [25, 82], norovirus [69], and Sars-Cov-2 [83, 84, 85], are subject to multi-epitope responses, although successful targeting of a single epitope is sufficient for neutralization [75]. However, the minimal model is insufficient to describe the full dynamics of immune recognition complexity

CAB

BAC

ACB

ABC

Selection 3

**A**: Table 3: Relative expression (RE) of genes in TLL1-transduced MT-2 cells over MT-2 cells transduced with an empty vector**B**: Genes were picked up if the RE of type 1 was larger than 11111111 and the RE of type 2 was smaller than 3333 or if the RE of type 1 was smaller than 1111 and the RE of type 2 was larger than 5555.**C**: Type 1(2) corresponds to MT-2 cells transduced with TLL1-isoform 1 (isoform 2)

BCA

ACB

CAB

Selection 2

**A**: (2019). Our study expands the concept of brain metastable states to multi-person interactions, providing a novel perspective without relying on millisecond time-scale synchronization**B**: Our work offers a fresh perspective on inter-brain connectivity by analyzing a hyperscanning neuroimaging dataset through multi-brain symbolic dynamics. Metastable states dynamics are crucial for brain flexibility, enabling seamless transitions between stable functional states and supporting cognitive processes and resilience to noise through robust mechanismsRabinovich, Zaks, and Varona (2020); Tognoli and Kelso (2014); Kelso and Tognoli (2006); Roberts et al**C**: Our results and analytical approach not only broaden the understanding of brain metastable states in the context of human interaction and coordination but also open new avenues for studying the brain mechanisms underlying these processes. This new tool bridges brain dynamics and behavior, offering a powerful and flexible method to explore the intricate neural choreography of multi-person interactions in a broad range of cognitive functions and collaborative behaviors.

BCA

BAC

ABC

CBA

Selection 2

**A**: The output special nodes and pair embedding of Co-Former are employed dependent on different tasks, including two pre-training tasks and two downstream affinity tasks. CN, PN, and RN are the special nodes for complex, protein, and RNA, respectively.**B**: Figure 2: Overview of CoPRA. Given a protein-RNA complex as input, the sequence information of protein and RNA are fed into a PLM and an RLM, respectively**C**: The output embeddings are selective with interface information and are fed into Co-Former with pairwise information. The Co-Former fuses the 1D and pair embedding by structure-guided multi-head attention and outer product modules, with a task-dependent attention mask

ACB

ABC

CAB

ACB

Selection 3

**A**: Artifacts in this dataset include out-of-focus areas, tissue folds, cover slips, air bubbles, pen marks, and areas with extreme over- or under-staining, which were carefully labeled to aid in quality control for pathology image analysis. We trained our network using 1500 artifact-free patches from epithelium and stroma classes. We then use this network to classify 474 artifact patches from 474 clean patches.**B**: HistoROI dataset is developed to segment WSIs into six key classes: epithelium, stroma, lymphocytes, adipose, artifacts, and miscellaneous**C**: For the experiment to filter out low-quality noisy image patches from good quality ones, we used HistoROI dataset [11]

ACB

CBA

ABC

BCA

Selection 2

**A**: Empirical validation of the urban scaling hypothesis has driven numerous studies using data from multiple countries and examining a wide array of urban indicators. In the realm of disease incidence, the seminal work by Bettencourt et al. reported that HIV/AIDS cases in the USA scale superlinearly with population size [9], a pattern also observed in Brazil by Antonio et al. [12]**B**: The authors suggest that the superlinear scaling observed in some infectious diseases likely reflects the increased number of contacts among residents in larger cities compared to smaller towns [14]. In contrast, the isometric and sublinear scaling of other infectious diseases may be due to the insufficient allocation of medical resources in smaller cities. Patterson-Lomba et al. [15] reported that sexually transmitted diseases scale superlinearly with the population of urban areas in the USA, even when controlling for socioeconomic variables. They also found that income inequality positively correlates with disease incidence, while educational level is negatively correlated.**C**: Rocha et al. [13] extended this analysis by investigating the scaling of various health indicators using data from Brazil, Sweden, and the USA. Their findings revealed that infectious diseases such as HIV/AIDS, chlamydia, and influenza generally scale superlinearly with population size. However, other infectious diseases, including leprosy, viral hepatitis, and dengue, exhibited isometric or even sublinear scaling

ABC

BCA

CBA

ACB

Selection 4

**A**: The results show the promising value of GAN-TAT, both through computational performance and validation against clinical data. However, inherent limitations of PIN embedding, such as imbalance, sparsity, and potential overfitting, could effect GAN-TAT’s efficacy. Despite these challenges, this work provided an innovative solution to this topic and offers a promising direction for continued research and enhancement. **B**: Utilizing the ImGAGN structure, GAN-TAT helps mitigate challenges associated with the use of PIN**C**: This study introduces GAN-TAT, a novel machine-learning framework that leverages network embedding for identifying druggable genes

ABC

CBA

CAB

ACB

Selection 2

**A**: In order to capture the hierarchical nature of GO labels, labels are represented as a directed acyclic graph and embedded as an additional feature input. They were able to achieve state-of-the-art results for 2 out of 3 problem subclasses and competitive results with NetGO and GOLabeler for the remainder. Notably, pre-training the Transformer through unsupervised training did not lead to performance improvements. **B**: [91] uses bidirectional LSTMs to condense input sequences into a fixed-size feature vector, which is then used to classify GO functional labels, although performance is not compared with [59, 58]. TALE uses the Transformer architecture [18] to perform GO label classification but with an additional label embedding**C**: More sophisticated architectures such as RNNs and Transformers have also been used. One of the challenges from dealing with protein sequences is the large variation in sequence length

ACB

BAC

CBA

Selection 4

**A**: VNN architecture and training**B**: The VNN model was pre-trained on a healthy population to glean information about healthy aging**C**: To facilitate ΔΔ\Deltaroman_Δ-Age that is transparent and methodologically interpretable, we used a multi-layer VNN model that yielded representations from the input cortical thickness features at the final layer, such that the unweighted mean of these representations formed the estimate for chronological age. Details on how this choice of architecture leads to anatomically interpretable ΔΔ\Deltaroman_Δ-Age are discussed subsequently.

ABC

BAC

BCA

Selection 1

**A**: Now, we discuss the conditions regarding the non-existence of spatially heterogeneous steady states**B**: For this purpose, first, we deduce a priori estimates for non-negative solutions of the system (17).**C**: The following two lemmas can be found in [37, 65], respectively

ABC

BAC

CAB

ACB

Selection 4

**A**: Building on this, Wu et al. (2023, CcBHLA) used BiLSTM, while TransPHLA (Chu et al., 2022; 2021) incorporated self-attention modules to capture complex dependencies**B**: Ye et al. (2023, STMHCpan) modeled peptide-MHC interactions as graphs, introducing graph neural networks to the field. **C**: With advances in natural language processing, sequence modeling techniques have gained popularity. NetMHCpan (Borole & Rajan, 2024; Reynisson et al., 2020; Jurtz et al., 2017) applied LSTM to both MHC and peptide sequences

ACB

BAC

BCA

CBA

Selection 3

**A**: The data generated and used within this work can be found at the Open Science Framework repository (Lluís Hernández-Navarro, Kenneth Distefano, Uwe C**B**: 2024. Supplementary data, code, and videos for “Slow spatial migration can help eradicate cooperative antimicrobial resistance in time-varying environments”. OSF. https://doi.org/10.17605/OSF.IO/EPB28).**C**: Täuber, and Mauro Mobilia

ACB

BCA

CAB

Selection 1

**A**: (2021) and how to identify the effect of ionization on receptor proteins. The examples reproduce existing results, confirming the reliability of our approach, but also demonstrate additional new discoveries made possible by systematic comparison of molecular simulations.**B**: (2020) how to quantify the influence of force-field parameter changes on simulations of nucleic acids,Cruz-León et al**C**: Here, we discuss the functionality included in PENSA and demonstrate its usefulness on three real-world applications: We show how to describe the influence of local frustration on loop opening during the catalytic cycle of an oxidoreductase,Stelzl et al

CAB

ABC

CAB

CBA

Selection 4

**A**: Figure 4 represents the scalp topographies of the grand average difference ERP**B**: We can see from figure 4 that the effect starts being significant from the timepoint 500 ms to 800 ms (p<0.001𝑝0.001p<0.001italic_p < 0.001) at the parietal region. In each topography, there is also another significant cluster at the frontal region which likely represents the negative polarity of the effect. **C**: The areas marked by black dots indicate significant differences in effects between target and non-target ERPs as determined by the spatial cluster permutation test described in 2.7, using the same threshold selection procedure as mentioned above

ABC

BAC

ACB

CBA

Selection 3

**A**: An extensive body of theoretical — and some experimental — work on proofreading reports the intuitive speed-accuracy trade-off. How might our work, motivated by experimental observations of stalling [33, 34, 35, 36, 45, 38], be consistent with such prior work? Prior studies consider only the impact of a few mutations or one biophysical perturbation, e.g**B**: changing Mg2+ levels [46]**C**: Such tests cannot probe the nature of any potential tradeoff between speed and fidelity. More precisely, trade-offs manifest as a Pareto front defined by an inequality between traits such as speed and accuracy.

CBA

BCA

ABC

CBA

Selection 3

**A**: DICOM pixel intensities were rescaled, and the image was resized to a smaller size of 224 x 224 to fit on the GPU. A pretrained Efficient-Net model was used for image processing, and extracted features were then concatenated with LSTM/Neural CDE’s features and passed through a final fully connected linear layer to give the final output.**B**: Similar preprocessing steps were performed on the structured data but for the multimodal data, each training step involved the incorporation of a random slice of the subject’s baseline CT scan**C**: To test the hypothesis for the improvement in both LSTM and Neural-CDE performance, a baseline CT scan image was used

ACB

CBA

CAB

Selection 3

**A**: Figure S11: The FI predicts future FPFP5 deficits better than NFPFP5. HRS, sex-stratified**B**: The curves are similar between sexes. Visually, individual points appear to agree between sexes within error**C**: The AUC is the probability that a metric will correctly rank positive individuals as higher than negative individuals [31] (0.5 is guess, dotted line; 1 is oracle). Note: the x-axes are sorted by AUC. Leave-one-out excludes the outcome deficit from the predictor (doesn’t affect FP frailty prediction). Error bars are standard errors.

BCA

ABC

CAB

Selection 3

**A**: (2023), which includes genomic sequences from known organisms—both from human genomes and a curated selection of genomes from multiple species (e.g., fungi, mammalian, invertebrate, bacteria)—and shuffle it into our metagenomic reads at a 1:8 ratio. **B**: For this, we sample sequences from the dataset provided by Zhou et al**C**: The modified training distribution aims to allow for us to maintain performance on metagenomic tasks, such as metagenomic embedding and classification, while also achieving improved performance on a broader set of genomic tasks (i.e., tasks involving non-metagenomic data)

CAB

CBA

BAC

BCA

Selection 2

**A**: Specifically, DNA methylation levels change rapidly during early development and adolescence, stabilize during adulthood, and may alter again in older ages [3, 4]. This nonlinearity[5] poses significant challenges for traditional age prediction models, which often assume a uniform rate of change across the lifespan**B**: However, modeling DNA methylation age prediction effectively is complex due to two intrinsic properties of CpG methylation. Firstly, the correlation between CpG site methylation levels and age is not static but changes with age, which we call Epigenetic Correlation Drift (ECD)**C**: Consequently, relying on a static overall linear correlation value can overlook CpG sites that are highly correlated with age in specific ranges, such as 40-50 years, but exhibit low correlation outside these ranges. Different CpG sites exhibit distinct methylation change patterns, called Heterogeneity Among CpGs (HAC). For example, a CpG site might have a high correlation with age during one age stage but a low correlation in the next, while other CpG sites maintain a high correlation in the subsequent stage.

BAC

BCA

CAB

BCA

Selection 1

**A**: Furthermore, we compute the reconstruction error as a measure of encoding quality (for details see Appendix E1) and observe that the continuous model is more unstable in terms of representation quality (Fig**B**: This may be due to a continuous drift of neuronal selectivity, which does not appear to be a feature of the discrete model (see Appendix figures F.7 and F.8). We conclude that a sufficiently long hold period and waiting for the stable state of the dynamics is important as recurrent dynamics remove the statistical dependencies and redundancies in the neural activity. This, in turn, allows learning factorised representations.**C**: 1D)

BCA

BAC

ACB

Selection 4

**A**: We demonstrate the wide applicability of UniGuide by tackling a variety of geometry-constrained drug discovery tasks**B**: With performance either on par with or superior to tailored models, we conclude that UniGuide offers advantages beyond its unification**C**: Firstly, while the novelty of conditional models often stems from the condition incorporation, our method redirects focus to advancing unconditional generation, which directly benefits multiple applications. Furthermore, this separation of model training and conditioning allows us to tackle tasks with minimal data, a common scenario in the biological domain.

ACB

BAC

ABC

BCA

Selection 3

**A**: DynamicBind predicts the ligand-specific protein-ligand complex structure with a deep equivariant generative model.**B**: NeuralPlexer (Qiao et al., 2023) incorporates essential biophysical constraints and a multi-scale geometric deep learning system for the diffusion process**C**: For generating the ligand-specific protein-ligand complex structure, a deep equivariant generative model named DynamicBind (Lu et al., 2024) is employed

CAB

CBA

ACB

Selection 1

**A**: SCN9A encodes the Nav1.7 sodium channel, which plays a crucial role in pain perception and anesthetic analgesia. CNR1 encodes the cannabinoid receptor type 1, influencing pain and the effects of cannabinoid anesthetics. SLC6A3 and SLC6A5 encode dopamine and glycine transporters, respectively, which are involved in neurotransmission and anesthetic mechanisms.**B**: Figure 4 presents the ADMET screening results for datasets involving GABRA5, SCN9A, CNR1, SLC6A3, and SLC6A5**C**: Specifically, GABRA5 encodes the α⁢5𝛼5\alpha 5italic_α 5 subunit of the GABAA receptor, which is targeted by benzodiazepines and barbiturates for sedation

CAB

BCA

BAC

CBA

Selection 1

**A**: We systematically trained the model to reconstruct images using sets of 4, 16, 32, and 64 learnable attention maps**B**: This approach allows us to evaluate the optimal number of features required for effective reconstruction**C**: The quality of the reconstructions is quantitatively compared with a baseline model, based on the end-to-end reconstruction model of Shen et al. [11]. Additionally, we assess the consistency of the spatial and feature inverse receptive fields (feature IRFs) across different stimuli by computing the standard deviation. Lastly, we employ a data-driven approach to visualize changes in feature RFs, using t-SNE for dimensionality reduction to explore whether electrodes exhibit preferences for certain features.

CBA

BAC

ACB

ABC

Selection 4

**A**: One great source of explanations is illusions and disorders of body representation. The human brain constantly sends and receives a flow of multisensory information**B**: We’ve already finished defining our model and we want to get a further step on rationality. For human beings, the most classical perspective we get self-conscious experience is our body**C**: The integration of this information is responsible not only for the way our body is represented but also for the way it is consciously experienced. Recent studies in cognitive neuroscience have shown that it is possible to modulate bodily self-conscious experience by experimentally changing these multisensory bodily signals. For example, Indian neuropsychologist Vilayanur Ramachandran designed an interesting experiment using a mirrored box to generate synesthesia and kinesthetic illusions in phantom limbs.

BAC

ACB

ABC

ACB

Selection 1

**A**: demonstrate their emergence within a relatively simple setup: a ring network of N𝑁Nitalic_N FHN units with symmetric connections but non-uniform parameters**B**: Contrary to the assumption that complex networks are essential for chimera states, Omelchenko et al**C**: The dynamics of each node in the network are described by:

BAC

ACB

ABC

ACB

Selection 1

**A**: 6, a simulation within a one-dimensional linear system illustrates the effect of a driving traveling pulse (panel A) - characterized by a constant velocity and width, and a heaviside-like profile - on an oscillatory system with a period of T≈186𝑇186T\approx 186italic_T ≈ 186**B**: In Fig**C**: The interaction results in phase patterning, as evidenced by a wavelength (λ≈1860𝜆1860\lambda\approx 1860italic_λ ≈ 1860) (panel B). Notably, the regions encountering the driving wave near the critical bottleneck zone exhibit pronounced phase differences, aligning with theoretical expectations (Panel C).

BAC

ABC

CBA

Selection 1

**A**: However, its limited reconstruction performance constrains its applicability. Lin et al**B**: More closely related to our work, Gao et al. (2023) adapts VQ-VAE (van den Oord et al., 2017) for protein structures**C**: (2023a) explores discrete structural representation learned by such VQ-VAEs (van den Oord et al., 2017), while Liu et al. (2023) trains a diffusion model on the discrete latent space derived from this approach. Similarly, and concurrently with our work, Liu et al. (2024a) combines finite scalar quantization (Mentzer et al., 2024) with a specialized transformer-based autoencoder for proteins, RNA, and small molecules.

BAC

BCA

ACB

CAB

Selection 1

**A**: A modified version [8] of the de Gruijl model [7] was implemented in Arbor-Pycat. The morphology of mouse IO neuron C4A was used [25]**B**: For programming ease, these default values were scaled by the parameter θ𝜃\thetaitalic_θ, which was thus a vector of ones, initially. Optimization was performed against the loss-function:**C**: After discretization, it contains 4 axonal, 15 dendritic and 2 somatic compartments. Initial parameters are listed in Table 1

ABC

ACB

CBA

CAB

Selection 2

**A**: Phages infect host cells by adsorbing (attaching) to receptors on the host cell wall and then delivering the genomic content into the host cytoplasm. Phages are much smaller than bacteria and each host cell presents multiple receptors that phages can bind to, so multiple phages can adsorb to a single host cell, though not all adsorptions necessarily lead to infection**B**: Multiple adsorptions become increasingly likely at higher phage densities (Turner and Duffy, 2008; Christen et al., 1990) and can become the dominant transmission mode at sufficiently high densities (Turner and Chao, 1999). If phage densities are very high, it is possible that multiple phages simultaneously adsorb to and then infect the same host cell**C**: Here, we explore the impact of simultaneous infections on phage-host ecology. We define simultaneous infection as infections that occur within a very small time window and distinguish between simultaneous infection and previously studied forms of co-infection, where after a pause an already infected host cell is infected again. Interestingly, given sufficient time phages can prevent multiple, sequential infections through host cell manipulations (Joseph et al., 2009) but these mechanisms are not applicable to the small time window relevant for simultaneous infections.

BAC

BCA

ABC

BAC

Selection 3

**A**: In this chapter, I have provided a brief introduction to neural networks consisting of DNA, using as an example the winner-take-all network proposed in Ref. Cherry and Qian (2018). The input data is provided as a DNA strand and is processed via biochemical reactions**B**: (2013). Such approaches constitute a promising starting point for the development of intelligent matter based on biological materials, and might also find applications in, for instance, medical contexts where input data is already present in a biochemical form.**C**: On this basis, it is possible to recognize handwritten digits using DNA. Moreover, I have briefly discussed a proposal for DNA-based reservoir computing Goudarzi et al

BCA

BAC

CBA

ACB

Selection 4

**A**: Positive correlations are observed in the temporal and parietal regions, indicating theta rhythms’ role in auditory information processing and memory integration**B**: The Text model shows reduced but still present positive correlations in the temporal lobe, highlighting the involvement of theta rhythms in cognitive processing of text-encoded auditory stimuli. For the Alpha band (8-12 Hz), the Audio model shows positive correlations in the occipital and parietal regions, consistent with alpha rhythms’ association with relaxed states and sensory processing. Negative correlations are prominent in the frontal cortex, suggesting active suppression of irrelevant information during auditory processing. The Text model exhibits a similar pattern with pronounced negative correlations in the frontal regions, indicating alpha rhythms are engaged during both auditory and text-encoded auditory processing, with significant involvement of cognitive control regions.**C**: The Theta band (4-8 Hz) reveals significant negative correlations in the frontal regions for both models, particularly in the prefrontal cortex, involved in working memory and executive functions during auditory processing

BCA

ABC

ACB

Selection 1

**A**: Even when the initial position of the agent is uniformly distributed throughout the system, the agent is not able to learn about its complete environment and gets trapped in nearest attractant peak, for low exploration rate and high learning rate**B**: When the attractant profile has peaks of different sizes, the exploration-exploitation trade off is even more pronounced**C**: As a result, even after a long time has passed, the agent may show same affinity for the lower and higher attractant peaks, which is detrimental for its performance. An optimum balance between exploitation and exploration helps the agent overcome the trapping effect and show its best performance. For non-uniform initial condition, when the agent starts near the lower attractant peak, its probability to localize near the higher peak after a long time, shows a maximum for intermediate rates of exploration and learning.

CAB

CBA

BAC

BCA

Selection 3

**A**: We plotted the correlation values for all the best performing models from conv1, conv2 and conv3 layers. Each point represents a signal-wise model, and the color represents the model type (blue is AFRT, red is Linear-AlexNet)**B**: Our analysis shows that the predicted activity from the AFRT model correlate higher with ground truth signals compared to the predicted activity from the baseline model Linear-AlexNet (Fig. 2)**C**: Although the baseline model makes use of a significant higher amount of features, our results show that models containing the affine components perform better (with a correlation value of 0.5 or higher). Overall, AFRT encodes MUA activity more accurately than the Linear-AlexNet model and our results also show that AFRT is less prone to overfitting.

BAC

ABC

ACB

CBA

Selection 1

**A**: Genes that have been previously shown to be downregulated in HD from which we obtain a negative correlation between SHAP and gene expression in SPN clusters are: Pde10A [19] or Scnb4 [20]. We have validated other lower rank genes previously described in different HD models, such as Penk1 (cluster iSPN), Gria3 (clusters iSPN, dSPN) [21] or Nrg1 [22]. Although our main focus was put on SPN clusters, other neuronal clusters are worth mentioning since they show shared altered genes with SPN clusters, eg. Gria3 and Nrg1 in clusters eSPN and adult NPC, Onecut2 in eSPN Adam12, Pcp4 in eSPN Rxrg and Adult NPC, or Pde10A in eSPN, Interneuron Whrn, Cholinergic Adrenergic neurons. **B**: Further model validation was made from SHAP results based on published data obtained from in vivo HD models or patients. For this validation, we compare the correlation between SHAP values and gene expression of previously described altered genes, expecting to observe a positive correlation between higher HD expression levels with positive SHAP values and a negative correlation between reduced HD expression levels with negative SHAP values**C**: Among the top 20 informative genes, some were previously described as being upregulated in HD. For them, we observe a positive correlation between SHAP and gene expression in SPN clusters are: Onecut2 [17] or Sfmtb2 [18]

BAC

CAB

BAC

BCA

Selection 2

**A**: For instance, the file parsing is highly optimized and executed with asynchronous buffers**B**: All data is read in streams, so that the number of input files, and their sizes, do not significantly affect the amount of required memory. Processor-intensive steps, such as file parsing and the statistics computations, are multi-threaded with a shared thread pool to leverage modern multi-core systems, and we paid close attention to selecting appropriate data structures for efficiency.**C**: Performance in runtime and memory was one of the major design goals

CBA

ACB

CAB

BCA

Selection 4

**A**: Our training data was obtained by extracting 50,000 pairs of conformations and molecules from this dataset**B**: The GEOM-QM9 dataset is an extension of the original QM9 dataset, enriched with multiple conformations for each molecule, primarily focusing on smaller compounds with a maximum of 9 heavy atoms**C**: For testing purposes, we reserved 17,813 conformations derived from 150 molecular structures.

CBA

BCA

ACB

BAC

Selection 4