Papers

RNA design aims to identify sequences that fold into specified secondary structures. Existing methods formulate the task as target-specific search or conditional generation. However, natural RNA evolution proceeds through sequence variation and selection, with compensatory substitutions, whereas these methods do not explicitly model this process. To address this limitation, we propose a two-stage framework comprising RNA Inverse-Folding Flow (RNA-IFlow) and RNA-IFlow-RL. RNA-IFlow uses structure-conditioned Dirichlet Flow Matching to model coordinated variation across the sequence, while RNA-IFlow-RL maps the learned flow to a pairing-preserving finite policy and refines it with thermodynamic feedback. Our framework achieves leading performance on multiple benchmarks, reaching 85.19% Pass@1 on Rfam-27. Further analyses reveal thermodynamic gains, policy dynamics, and robustness across settings. Our work couples coordinated variation with thermodynamic selection, offering a novel paradigm for RNA design.

Generative models make molecular crystal structure prediction fast, but their samples still exhibit geometric and packing violations. Physics can be introduced during training, post-training, or inference, yet these choices are rarely compared with the generator and physical signal held fixed. We introduce CrystAF, an all-atom crystal flow-map generation model, and use it with the UMA interatomic potential to systematically study where physics should enter. Post-training learns physical preferences directly into CrystAF, improving molecular validity and crystal packing while leaving sampling unchanged: physics is paid for once during training rather than repeatedly at deployment. In contrast, UMA relaxation is effective at repairing local clashes but makes generation 6–26$\times$ slower, while learning from relaxed targets provides little benefit. These routes are complementary rather than competing. Physics-informed post-training first shifts the generated distribution toward more physically reasonable structures, after which inexpensive inference-time corrections further remove clashes and restore stereochemistry that the generator cannot represent. Importantly, the same post-training strategy also improves the multi-step all-atom Clari-M and rigid-body MolCrystalFlow generators, demonstrating transfer across architectures and representations. Together, our results suggest a simple principle: learn reusable physical alignment into the generator, and reserve inference-time physics for residual constraints that are better corrected than learned.

Deep learning models have achieved strong performance in artificial intelligence for science, yet their black-box nature limits our understanding of how they learn scientific tasks. Existing methods for interpretability provide limited insight into how models organize evidence and evolve during learning. We introduce explainability from training (EFT), a model-agnostic paradigm that traces model interpretation during training to explain why models rely on specific features and how they organize these features as predictive evidence. We apply EFT to four state-of-the-art T cell receptor (TCR)-epitope prediction models, TCR-SRIM, TULIP, MixTCRpred, and NetTCR-2.2, spanning post-hoc and interpret-by-design approaches as well as transformers and CNNs. To investigate how structural information affects model explanations, we introduce a benchmark, TCR-XAI2, containing 388 unique experimentally resolved TCR-epitope structures, complemented by structures predicted using AlphaFold3, Boltz-2, TCRModel2, tFold-TCR, and OpenFold3. Using EFT with TCR-XAI2, we demonstrate that (1) CNN and transformer models exhibit distinct learning trajectories; (2) TCR $\alpha$ and $\beta$ evidence can conflict during learning, limiting the benefits of jointly modeling both chains, while MHC information mitigates this; and (3) real versus predicted structural data for TCR-epitope prediction exhibits distinct TCR and peptide feature preferences as well as differing trajectories of model certainty.

Biological machine learning was long bottlenecked by the ability to synthesize designed DNA. Variational synthesis models control chemical reactions to physically manufacture quadrillions of designed sequences in DNA. However, training these generative models is challenging: constraints on chemical synthesis can force many parameters into a discrete space, limiting the ability to pre-train and fine-tune. In this article we train “free” variational synthesis models using stochastic gradient descent in continuous space, and then discretize with post-training quantization to impose hardware and wetware constraints. This enables variational synthesis models to satisfy stringent reward criteria, while still synthesizing diverse designs, achieving a strictly dominating quality-diversity Pareto frontier. We demonstrate by training variational synthesis models of enzymes, peptides, antibody CDRH3s, and regulatory DNA elements. In silico performance is maintained in vitro.

Recent high-resolution (500 bp) Hi-C experiments reported contact maps with unusual patterns around highly expressed bacterial DNA loci. These patterns, described as arched stripes and bundled domains, extend over nearly 100 kbp. We performed Brownian Dynamics simulations with a specially designed coarse-grained model to rationalize these findings. The main feature of the model is that it takes explicitly into account the waves of positive (respectively, negative) supercoiling generated by the translocating polymerase downstream (respectively, upstream) of its position. Contact maps computed from the simulations also display the arched stripe and bundled domain patterns for above-threshold values of the rate of twist injection. Computed DNA conformations indicate that these patterns reflect an extraordinarily entangled DNA geometry, which involves both plectonemic and toroidal supercoiling, with some DNA segments experiencing both of them simultaneously. Moreover, DNA segments located on each side of the polymerase systematically wind around tracts located on the other side, which is a purely out-of-equilibrium effect driven by the continuous injection of twist. The present work therefore reveals that repeated transcription of a DNA locus systematically brings into contact DNA segments separated by several tens of kbp, which may eventually contribute to allosteric modulation and long-range communication.

Small-molecule optimization integrates medicinal-chemistry reasoning and computational evidence through iterative, multi-objective decisions. When large language models (LLMs) reason over optimization histories stored primarily in conversational context, they must recover candidate identities, prior evaluations, and task constraints to guide subsequent decisions. We present M3OS, a multi-agent LLM system that decouples molecular-design reasoning from optimization-state management through Monte Carlo graph search. A persistent graph links evaluated candidates, parent-child transformations and evaluation evidence, while rewards and visit statistics guide LLM-assisted parent selection. Two branches combine tool-driven candidate generation with knowledge- and case-guided medicinal-chemistry editing. An execution harness controls graph updates through structured output extraction, molecular validation and task-bound evaluation. Agents receive role-specific contexts, while the graph preserves optimization trajectories beyond their active contexts. Across three molecular optimization benchmarks, M3OS achieves higher success rates than baselines, supporting the integration of persistent search state, specialized agents and controlled execution for multi-constraint optimization.

Proteins change shape as they function, yet most inverse folding models predict amino acid sequences from a single, fixed backbone. A central challenge in protein engineering is to design proteins that undergo specific motions, which requires accounting for how their structures change over time. This motivates inverse protein folding conditioned on protein motion. We introduce SPINET, which predicts sequences from molecular dynamics trajectories. It uses cellular sheaves to represent residue interactions within each frame and recurrent units to integrate information across frames, then predicts all amino acids in a single pass. We evaluate SPINET on mdCATH and ATLAS, where it outperforms all evaluated static and ensemble baselines in sequence recovery. On mdCATH, it achieves 56.7% top-1 recovery, compared with 44.5% for the strongest static baseline and 40.7% for the strongest ensemble baseline. We also evaluate whether the predicted sequences are compatible with conformations sampled along the target trajectory. On mdCATH, they achieve a median TM-score of 0.760, and structural recovery favors target conformations over unrelated decoys for 99.5% of test domains.

Protein–ligand binding affinity prediction is fundamental to computational drug discovery, yet modern AI-driven models are limited by pervasive heterogeneity in their training data: bioactivity values are aggregated across diverse assay types and experimental conditions without accounting for protocol-level differences, introducing systematic noise. Existing harmonization approaches either discard assay-level metadata or collapse it into coarse categorical distinctions, leaving rich contextual signal unused. We address this gap with two contributions. First, we introduce Assay-Aware BindingDB, augmenting 74,425 BindingDB protein–ligand pairs across four assay types (ITC, SPR, RBA, and FPA) with structured metadata extracted from primary literature using a two-stage agentic framework that separates evidence extraction from ontology-conditioned JSON synthesis. Against domain-expert references, the framework achieves presence F1 $\geq 0.947$ and semantic content accuracy $\geq 0.913$ across all assay types. Second, we develop a context-conditioned affinity model that injects an assay-context embedding into the Boltz-2 affinity module. On a paper-level held-out split, the model reduces combined-regime MSE from $1.27$ to $1.19$ and increases Pearson correlation from $0.64$ to $0.67$, with significant gains for SPR and FPA. ITC, the only label- and immobilization-free assay considered, shows no improvement, consistent with its design removing the protocol artifacts the metadata captures. These results support the hypothesis that systematically curated assay metadata provides informative signal for affinity prediction.

Persistent homology and persistent Laplacians have become effective descriptors of large molecular binding interactions, the latter encoding multiscale geometry and spectral information beyond the harmonic subspace. Both, however, are built on undirected graphs and pairwise contracts, while the interactions that determine binding are often directional and involve more than two atoms at once. We introduce persistent hyperdigraph learning (PHL), which represents these directed, many-body interactions directly. Because hyperdigraph Laplacian spectra become costly to compute at higher topological orders, we further use stochastic trace estimation to develop a matrix-free formulation(MFPHL) that estimates Laplacian trace statistics from probe-based quadratic forms evaluated through sparse boundary operators, without assembling or diagonalizing. The number of probes required for a target relative accuracy of the trace estimate is independent of matrix dimension, and per-probe cost scales with operator sparsity. On protein-protein binding affinity benchmarks, MFPHL reduces feature-generation time by roughly two orders of magnitude relative to the eigenvalue-based pipeline. Its predictive accuracy matches that of spectral PHL on the P2P wild-type set within training-seed variability but is lower on the two larger datasets. This dataset-dependent trade-off brings higher-order hyperdigraph features within practical reach.

Although $\mathbb{E}[\mathrm{dropout}(x)] = x$, here we show that $\mathbb{E}[\mathrm{LayerNorm}(\mathrm{dropout}(x))]$ is not equal to $\mathrm{LayerNorm}(x)$. Accordingly, the pattern of a Dropout layer followed by a LayerNorm, which is common to many AlphaFold2-based protein structure predictors, produces a systematic bias at evaluation time that can hamper performance. To address this, we derive a closed-form, first-order correction for this gap, which we call a Dropout-LayerNorm Correction (DLC). DLC empirically matches the performance boost of large Monte Carlo dropout ensembles. We evaluate its effect across nine protein structure models (ESMFold, OpenFold, ABB3, FlashABB, Ibex, NbForge, Genie1, Genie2, Genie3) on both paired and single-chain antibody structures as well as one protein-ligand docking model (QuickBind). The correction is computationally negligible and improves accuracy in all ten models tested ($\sim0.3\%-13\%$), with a modest but consistent improvement in ESMFold and OpenFold and a substantial improvement in antibody-specific models. This work identifies the mathematical consequence of chaining together Dropout and LayerNorm and provides a free, principled adjustment to improve the evaluation performance of many pretrained models. Code to reproduce the experiments is available at https://github.com/oxpig/DLC.

Optimizing atmospheric pressure interfaces requires balancing efficient ion transport with unwanted collision activation. The complex environment at the inlet region, including rapid pressure drops, RF/DC electric fields, ion-gas collisions, and space charge effects, makes simulation of this system challenging. In this work, we introduce MuCITE (Multiphysics simulation for Continuous Ion Transport and Energetics), which combines continuous ion injection, precomputed electric and gas flow fields, and particle-in-cell space charge updates. Furthermore, by employing improved impulsive collision theory, it can not only predict transport but also analyze the loss position of ions and the evolution of their kinetic energy, internal energy, and dissociation probability during transport. Transport predictions with ion currents and activation trends of in-source fragmentation were compared with experimental measurements using a custom-built source and a Q-Exactive HF source. The results show that the simulation predictions agree well with the experimental results and can serve as a reference before experiments. MuCITE allows users to derive ion transport, output beam characteristics, and ion energy through the configurable interface geometry and operating conditions. It supports simulation-guided design and tunable parameters to achieve predicted efficient ion transport or in source activation.

Recent advancements in image-based profiling techniques have enabled the collection of high-volume cell morphology data, allowing new molecular embedding models to learn from the experimental phenotypic perturbations of a molecule in a cell. Previously, we developed Molecule-Morphology Contrastive Pretraining (MoCoP), a strategy for aligning small molecule embeddings to morphology fingerprints extracted through CellProfiler. The resulting molecular representation showed transferable performance for quantitative structure–activity relationship (QSAR) prediction tasks. Here, we extend the method by using a deep-learning-based cell image encoding pipeline to extract more feature-rich morphology profiles and align them to the molecular embeddings through contrastive learning. The new embeddings encode more accurate information on how molecules perturb cell morphology and enable improvements for QSAR predictions through either fixed-embedding linear probes or fully flexible fine-tuning. Morphology retrieval performance scales log-linearly with training data size, suggesting continued improvements as larger datasets become available. The improved MoCoP v2 also achieves superior performance on toxicity prediction and competitive results on ADME and activity benchmarks, when compared with existing molecular embedding models that use both cell morphology and transcriptomic data during training.

Single-molecule affinity mapping enables measurement of individual proteins and proteoforms, but imperfect and nonspecific probe binding means that affinity traces may be compatible with multiple proteoforms. Accurate abundance estimation therefore requires probabilistically weighting ambiguous traces rather than assigning each trace to a single candidate. We developed ProteoEM, an expectation-maximization framework for weighted proteoform quantification, inspired by transcript abundance estimation methods for RNA-seq and released as an open-source Python package. ProteoEM evaluates each observed affinity trace against all candidate proteoforms using fixed, pre-calibrated probe-response rates that are separate from abundance estimation. It estimates the abundance of each proteoform and reports proteoforms that the probes cannot tell apart as a single group. Because some proteoforms are observed more readily than others, ProteoEM also corrects for these observation yields to estimate the composition of the source sample. We show that grouping traces that carry the same evidence into equivalence classes reduces the EM's work sevenfold without changing the estimates. In simulations, ProteoEM recovered the true molecular composition, whereas approaches that reduced each affinity trace to a hard yes/no call introduced substantial errors. Performance was robust to moderate, uniform calibration error but was biased by informative missing data and by proteoforms absent from the reference set. When observation yields were known, ProteoEM also recovered source-sample composition from observed molecular counts. ProteoEM is an open-source, reproducible framework for quantitative analysis of single-molecule affinity measurements and provides a basis for validation using experimental molecule-level data.

Protein modification requires navigating an immense sequence space, yet wet-lab validation remains low-throughput and costly. Although computational paradigms including protein language models (PLMs), large language models (LLMs), and LLM-based agents have shown promise in protein modification, their relative efficacy across realistic experimental decision-making settings remains unclear. To bridge this gap, we introduce PFArena, a benchmark comprising four controlled task interfaces that cover single-mutant generation and multi-mutant ranking. By providing varying levels of mutation fitness data, PFArena reflects four representative research scenarios characterized by differing degrees of prior experimental context. We assess six PLMs, six LLMs, and five LLM-based agents using complementary metrics to measure both peak and overall protein modification performance. Our evaluation reveals that model performance shifts systematically with the availability of target-specific experimental evidence: PLMs demonstrate proficiency in open-ended single-mutant generation by leveraging protein-specific priors, whereas LLMs and agents perform strongly in multi-mutant ranking, particularly when target-specific fitness data are available. Nevertheless, all model families face fundamental challenges with increasing search-space size and mutation depth. We release our code and benchmark suite to facilitate reproducible research in model-assisted protein modification.

Kinetic models of protein aggregation describe populations by size, not by spatial organization or morphology. We extend Lumry-Eyring nucleated polymerization to a model in which the monomer is a density field and each aggregate is a region bounded by a level set. Growth is a flux condition on the available sites of a surface. Condensation is a reaction between the bonding sites of two surfaces in contact, at a rate set by the bond rate and the contact geometry. The availability of those sites is a field on the interface, and its equilibrium value follows from Wertheim's perturbation theory. The collision efficiency and the Fuchs stability ratio are therefore computed, not fitted. In a well-mixed limit the model's spatial averages satisfy the rate equations term by term; the monomer fraction agrees to eight parts in ten thousand, a difference that arises from equating aggregate size with volume. The condensation kernel's exponent is $0.5806\pm0.0013$ against the $0.600\pm0.010$ fitted to a monoclonal antibody. The computed stability ratio reproduces thirteen of fourteen published conditions at twelve $k_BT$, but only with the bond rate at the top of its range. In a many-body box, aggregates merge at $2.2$ to $3.8$ times the two-body rate.

Modern imaging techniques can resolve individual pathological protein aggregates in postmortem human samples, providing detailed measurements of aggregate size distributions that are inaccessible with conventional bulk approaches. These distributions represent mechanistic fingerprints of the microscopic processes that generated the observed pathology, but extracting this mechanistic information requires a quantitative theoretical framework. Here, we develop the mathematical tools needed to interpret aggregate length distributions in living systems, where aggregate growth competes with active removal. We show that, across a class of models, the length distribution of sufficiently large aggregates approaches a geometric decay. Crucially, the decay rate is determined by the balance between aggregate elongation and removal, providing a direct quantitative readout of these competing processes from a single time point measurement. This enables mechanistic comparisons between healthy and diseased human samples without requiring longitudinal measurements of aggregate dynamics. We further analyse how additional aggregation and removal processes modify the observed length distributions. Together, these results establish the mathematical foundations and tools to use aggregate length distributions as an experimentally accessible route for inferring microscopic aggregation dynamics directly from human tissue.

Medical image analysis remains fundamentally challenging because of the intricate geometric and topological structures present in medical data. Conventional convolutional neural networks model images as regular Euclidean grids, limiting their ability to preserve geometric relationships and higher-order structural information. Recently, manifold topological deep learning (MTDL) has emerged as a promising paradigm that integrates deep learning with geometric and topological representations. Nevertheless, existing methods have not yet fully exploited discrete manifold structures within combinatorial complex neural networks. To bridge this gap, we introduce CNMTDL, a MTDL framework that integrates Hodge decomposition with a combinatorial attention mechanism. In our approach, medical images are represented as discrete manifolds and decomposed into three Hodge components. Features extracted from these components are concatenated and embedded into a combinatorial complex architecture, enabling enhanced higher-order message passing between $0$-cells and $2$-cells through attention-based blocks. We evaluate CNMTDL on six two-dimensional and three-dimensional datasets from the MedMNIST v2 benchmark, demonstrating its effectiveness for medical image analysis.

Nucleosomes are structures made up of proteins called histones that bind and compact DNA, driving the mesoscale organization of the chromatin polymer. Although histone proteins have diversified over evolutionary time, their contributions to the corresponding diversification of chromatin structure are poorly understood. Here, we mine protein databases for histones and create in silico nucleosomes for 3241 organisms spanning $>$1.5B years of evolution. Using a combination of electrostatic calculations and coarse-grained molecular dynamics simulations, we reveal extensive biophysical diversification of the nucleosome unit. Finally, we perform coarse-grained oligonucleosomal simulations on a subset of evolutionarily and biophysically divergent nucleosomes, demonstrating dramatic differences in bulk phase behavior of chromatin. Taken together, our results suggest a paradigm in which histones may have evolved to facilitate particular types of mesoscale chromatin behavior.

Proteins can possess numerous mutations relative to their wild-type amino acid sequences with minimal impact to their structure and function. However, in other cases, even a single amino acid mutation relative to the wild-type sequence can lead to a large change in structure or even a disease phenotype. While the accuracy of wild-type protein structure prediction has improved significantly in recent years, it remains difficult to accurately predict the structure of mutant proteins. Here, we characterize the local mutation-induced structural changes in proteins for a dataset of wildtype and the corresponding single-amino acid mutant x-ray crystal structures from the Protein Data Bank (PDB). We find that mutation-induced structural changes in these proteins are localized at the site of the mutation, decaying rapidly with increasing spatial distance from the mutation site. In addition, we evaluate how well AlphaFold3 can recapitulate the observed mutation-induced structural deformations in the x-ray crystal structures. We find that the accuracy of the AlphaFold3 predictions decreases strongly with increasing mutation-induced deformation. In contrast to the results for AlphaFold3, the Pearson correlation between a single physical feature, i.e. the change in solvent accessibility, and the mutation-induced deformation does not depend on the magnitude of the deformation. Our results and analyses provide a framework for further studies aimed at predicting the structural changes in proteins caused by single amino acid mutations.

Virtual screening aims to prioritize active compounds from large chemical libraries within a limited experimental budget. When applying Boltz-2 to virtual screening, a key challenge is how to use limited experimental data from the target assay to improve the prioritization of active compounds. We investigated whether fine-tuning the Boltz-2 affinity heads with a small number of binary activity labels could improve early enrichment of active compounds in hit discovery. We compared fine-tuning with 40-300 labels in a retrospective evaluation on eight MF-PCBA targets. With 300 activity measurements, fine-tuning increased the number of actives in the top 1% by a geometric mean of 1.77-fold across the eight targets and improved average precision (AP) by 2.14-fold relative to the control without fine-tuning. We also investigated whether rescoring a subset of candidates could retain the improvement in hit recovery by reranking only the top-ranked Boltz-2 candidates with the fine-tuned head. Restricting rescoring to approximately 10% of the evaluation set retained hit recovery comparable to full rescoring. These findings show that affinity-head fine-tuning with limited activity labels improves early enrichment with Boltz-2 and that this benefit can be retained when rescoring a restricted set of candidates.

The enzyme turnover number (kcat) defines catalytic efficiency and constrains quantitative models of metabolism, yet the molecular determinants governing kcat and its response to mutation remain poorly understood. Measurements are sparse and labor-intensive, and most computational approaches provide numerical predictions without explaining how enzyme-substrate interactions shape catalytic outcomes. A central challenge is therefore to identify the topological principles that determine where mutations act and how their functional outcomes are encoded within the enzyme-substrate interaction network. Here, we show that catalytic efficiency and mutational effects can be interpreted through enzyme-substrate interaction topology. We developed Interkcat, an interpretable bidirectional cross-attention framework that captures reciprocal coordination between protein residues and substrate atoms. Optimized on a unified benchmark, Interkcat achieves state-of-the-art predictive performance (R2 = 0.701). From its learned representations, we derive an Interaction Topology Score (ITS) that identifies sequence regions statistically enriched for mutation-sensitive sites without explicit structural inputs. We further demonstrate that higher-order topological features distinguish opposing mutational outcomes: lethal mutations disrupt coordinated networks, whereas activity-preserving or enhancing mutations retain sparse, globally organized coupling. These findings establish interaction topology as a unifying principle linking enzyme sequence, catalytic efficiency, and evolutionary perturbation.

Modeling protein-protein interactions requires accurate scoring functions that can rank potential poses (conformations) of a protein-protein complex to differentiate near-native poses from incorrect ones. Here, we propose a general framework for improving protein-protein pose ranking and other biomolecular interaction models using the LambdaLoss loss function from the Learning-to-Rank field. We test this framework by fine-tuning the energy prediction head of DFMDock with the LambdaLoss on an augmented dataset of 2.9M decoy poses derived from the DIPS dataset. On targets from the CAPRI score set benchmark, our fine-tuned ranking model LambdaDockScore is better at identifying correct poses in its top-1 and top-5 predictions compared to EuDockScore, a state-of-the-art method. LambdaDockScore also improves upon baseline DFMDock ranking performance for scoring antibody-antigen complexes and protein-protein complexes with very large or small binding interfaces.

Excess sodium intake remains a major public health challenge, while salty and saltiness-enhancing peptides offer a potential route to preserve sensory saltiness in reduced-sodium foods. Machine-learning studies of salty peptides, however, are constrained by small datasets, heterogeneous evidence standards, uncertain negative labels, and sequence similarity leakage. Here we present SaltyMeta, a curated benchmark and web-accessible screening framework for salty or saltiness-enhancing short peptides. The benchmark contains 580 peptides, including 280 positive peptides and 300 negative peptides, all standardized to 2-15 residue one-letter amino-acid sequences. Quality control found no non-standard residues, exact duplicates, or positive-negative overlaps. A similarity-grouped split retained 456 peptides for training and 124 for held-out testing. We evaluated 548 interpretable peptide descriptors, frozen ESM2 embeddings at 8M, 35M, and 150M parameter scales, and descriptor-embedding fusion models under grouped cross-validation. The traditional ExtraTrees baseline selected by training-set grouped cross-validation achieved ROC-AUC=0.693 in cross-validation and ROC-AUC=0.704, PR-AUC=0.702, F1=0.626, and MCC=0.304 on the held-out test set. The best initial protein-language-model fusion was traditional descriptors plus ESM2-8M embeddings, with grouped CV ROC-AUC=0.696 and test ROC-AUC=0.700. Advanced optimization using PCA95 dimensionality reduction and ExtraTrees feature-importance filtering yielded a practical ESM2-8M PCA95 top-300 model with test ROC-AUC=0.715 and PR-AUC=0.703, although repeated-CV gains remained modest. SaltyMeta is a transparent prioritization tool, not a sensory validation substitute. We provide benchmark, models, scripts, GitHub, and Streamlit for reproducible screening of food-derived peptides.

Ligand-based virtual screening depends on choices that are often treated as implementation details, including the similarity threshold, fingerprint setting and atom-invariant scheme. We examined how these choices altered the composition of ranked searches against the Enamine library for four aminergic reference ligands: atomoxetine, bupropion, mirtazapine and venlafaxine. Candidate sets were evaluated by compound-weighted scaffold novelty, normalized Shannon scaffold diversity and three computational estimates of synthetic accessibility. We first identified ligand-specific operating points along cumulative Tanimoto-ranked searches. We then compared five extended-connectivity fingerprint settings at matched retrieval depths and compared ECFP4 with the feature-class analogue FCSFP4. Finally, 240 records, corresponding to 229 unique structures, were scored independently by three chemists who were blinded to the computational scores. No single Tanimoto cutoff described all four searches. ECFP8 provided the most stable pooled setting, although venlafaxine favored ECFP2. ECFP4 was the stronger primary fingerprint in pooled comparisons, whereas FCSFP4 contributed nonredundant chemical space. Agreement among individual chemists was moderate, and the mean rating was more reliable than a single rating. SCScore showed the highest association with the blinded consensus, but performance varied by ligand. These results support a staged screening design in which a pooled default is followed by ligand-specific calibration and expert review.

Co-folding models such as AlphaFold3, Protenix, ESMFold2, and OpenDDE have advanced rapidly, yet no single model consistently performs best across all biomolecular complexes. In this paper, we show that their pair representations encode complementary information that can be transferred across models to improve structure prediction. We introduce SoupFold, which combines pair representations from multiple co-folding models in a common representation space and generates structures from the combined representation. Importantly, SoupFold does not retrain the co-folding models and learns only simple mappings to transfer representations across models. We evaluate SoupFold on antibody-antigen, protein-protein, protein-ligand, molecular glue, GPCR, and oligomeric complex prediction using AlphaFold3, Protenix, ESMFold2, and OpenDDE. By combining representations across models, SoupFold improves over individual co-folding models across the considered benchmarks.

Scientific research is a continuous process that emphasizes inheritance. Methods developed by predecessors are often expanded upon by new researchers to explore more novel and in-depth scientific questions. However, the change of lab staff, such as student graduation, leads to a lack of personnel capable of replicating methods. Methods that have been developed with significant effort and resources cannot be continued. To address these limitations, we propose LabAgent, a reproduce and discovery harness tailored for a lab's continuous work. LabAgent employs two mechanisms to guarantee that all skills can be executed and verified and to record the corrective methods and experiences, allowing for direct correction or avoidance of similar errors. We applied LabAgent to drug property prediction, biomedical problem analysis, protein variant effect prediction, and statistical genetics in life science domains. LabAgent ranks first over commercial generalist agents in every domain, and demonstrates accurate reproduction of a published figure. Overall, these results demonstrate that LabAgent can effectively integrate and reasonably expand laboratory knowledge.

Collision cross section (CCS), derived from ion mobility mass spectrometry, is a common descriptor for molecular annotation. Prediction is challenging for machine learning models because it reflects the size, shape, and ionization state of a gas-phase molecular ion. Most predictors either ignore explicit 3D structure or treat adduct identity as a late categorical feature, which limits their ability to capture adduct-dependent geometric effects. We present GRACE (Geometric Residual Adduct Conditioning via Early-fusion), a 3D CCS predictor that adapts a pretrained molecular geometry encoder using geometric residual adduct conditioning via early fusion. GRACE combines two inductive biases: a residual objective relative to an adduct-aware physical descriptor baseline and adduct conditioning within the encoder via a learned adduct token and low-rank attention adapters. We evaluate the model on a curated set of over 9,000 experimental molecule-adduct CCS records with random, scaffold, and adduct-sensitive splits designed to separate interpolation, scaffold generalization, and adduct-driven generalization. GRACE achieves the best mean percentage difference among the evaluated learned models on all three splits: 1.67% on the random split, 2.11% on the scaffold split, and 2.36% on the adduct-sensitive split. Diagnostic analyses suggest that residual learning stabilizes training by removing the dominant mass-CCS trend, while early fusion improves adduct-sensitive prediction relative to late fusion. Across four independent external test sets, GRACE shows consistently lower error than the other evaluated models. On a held-out set, GRACE also attains the lowest mean percent difference when compared with four previously reported physics-based workflows. These results support residual learning and encoder-level adduct conditioning as practical inductive biases for fast, accurate CCS prediction.

Computational RNA structure pipelines generate many candidate conformations for the same sequence. Reliable evaluation therefore requires more than recognising plausible geometry, it requires determining whether that geometry is compatible with the sequence. We introduce SIRGE, a sequence-informed geometric evaluator that conditions structural representations on nucleotide embeddings from a pretrained RNA language model. Early results show that SIRGE outperforms established evaluators in Kendall–$\tau$ alignment, Top-1 selection, and Top-3 ranking. Controlled comparisons further show that sequence conditioning corrects errors made by an otherwise matched geometric model and improves target-level rank structure. These findings provide initial evidence that pretrained sequence representations supply ranking information that complements geometric reasoning.

Cancer remains a major global health concern due to chemotherapy resistance and toxicity from high-dose treatments. To overcome these challenges, new therapeutic strategies targeting key proteins in cancer progression are essential. This study evaluates two phytochemicals, Carpaine (Car) and Rutin (Rut), from Carica papaya leaves, for their potential in enhancing cancer therapy by targeting B-cell lymphoma 2 (BCL-2) and WW domain-containing protein 1 (WWP1) proteins. We assessed their additive, allosteric, and synergistic effects using molecular docking, multi-ligand simultaneous docking (MLSD), molecular dynamics (MD) simulations, and MMPBSA analysis. Car and Rut showed an additive effect on BCL-2 by binding at distinct regions within the same pocket. MLSD revealed an improved binding affinity of -13.13 +/- 0.08 kcal/mol, compared with individual ligands or the commercial inhibitor Venetoclax. For WWP1, Car bound near the H-site and Rut near the Le-site, exhibiting an allosteric effect that increased Car's binding affinity in MLSD to -15.59 +/- 0.39 kcal/mol. Furthermore, Rut combined with bortezomib (Bort) demonstrated a synergistic interaction with WWP1. Binding energies were -7.64 +/- 0.156 kcal/mol for Bort, -10.26 +/- 0.07 kcal/mol for Rut, and -15.59 +/- 0.39 kcal/mol for MLSD, suggesting a more stable complex through synergy. These results suggest Car and Rut, particularly in combination with Bort, as promising candidates against cancer-related proteins BCL-2 and WWP1. Further experimental validation is warranted to explore their therapeutic potential.

Predicting protein dynamics is a long-standing problem in computational structural biology. Often, protein function critically depends on local directed motions, such as hinge movements, catalytic loop rearrangements and domain reorientations, which can be characterized by directional flexibility and correlated structural motions of the protein backbone. While Molecular Dynamics (MD) simulations provide an established but often prohibitively expensive approach, recent deep generative models aim to reduce this cost by directly predicting conformational ensembles, emulating MD. However, due to their large size and the need to generate several states until the derived dynamical properties converge, these models remain expensive. In this work, we propose BackFlip-2: a fast SE(3)-equivariant graph neural network trained to directly predict dynamical descriptors, such as directional backbone flexibility and pairwise dynamic correlations, from an equilibrium structure. In a series of experiments, we show that our model matches the accuracy of substantially larger ensemble generation models while being orders of magnitude faster, and demonstrate that the proposed equivariant architecture is especially well-suited for capturing anisotropic motions in proteins. BackFlip-2 model weights, training and inference code are available at https://github.com/graeter-group/backflip.