bioinformatics engineers

Abstract: Large-scale scientific workflows run on distributed HPC resources, making it hard to explain how scientific data products are generated and why similar workflow executions can yield very different results. Existing systems usually focus on either distributed workflow execution or provenance management and rarely treat fine-grained provenance as an intrinsic property of distributed scientific data artifacts. We present TRACEs (Traceable Runtime for Advanced Containerized Executions), a provenance architecture for distributed containerized scientific workflows that embeds data lineage and execution context directly in every generated data artifact. By decoupling provenance from workflow orchestration, TRACEs generalizes artifact-centric provenance from local container runs to distributed workflows while remaining compatible with existing workflow management systems. We evaluate TRACEs on a seven-stage E. coli variant-calling workflow distributed across multiple HPC nodes. Our results show 100% provenance completeness under distributed execution, about 59% less intermediate storage via sparse IMG containers, and runtime and zero-copy I/O throughput comparable to fixed-size SIF containers, plus provenance-guided explanations of workflow variability that trace a 200-fold difference in variant counts to differences in input sequencing data without re-executing the workflow.

Abstract: Single-cell assays study the genome one cell at a time. Unimodal assays measure one molecular layer, the newer multimodal assays measure several. Most data are single-modality, so integrating them relies on models trained on multimodal assays, where the cell correspondence across modalities is known. These models return a matching, but rarely a confidence measure. We present BASILISC (Bayesian Alignment of Siamese-Integrated Latents for Inferring Single-cell Correspondence), a framework for matching single-cell representations with uncertainty on the correspondence itself. A BASILISC model is a probabilistic siamese network over the two modalities with a predictive distribution over the correspondence for a cross-modal pair of cells. We instantiate it at two levels: at the cell-type level the question is whether two cells share a cell type, at the cell level which cell is the partner in the other modality. To our knowledge, this is the first uncertainty score on a learned cross-modal correspondence evaluated against a known ground-truth pairing. BASILISC learns a similarity across modalities that reflects cell lineage: an unseen cell type is matched to its lineage relative, matching still works on such held-out types (balanced accuracy 0.86 on unseen against 0.93 on seen types), and the predictive uncertainty rises on these out-of-distribution cells. The uncertainty score also selects high-quality matches for downstream analysis: discarding the least certain fifth of the pairs raises accuracy from 0.93 to 0.98. At the cell level, a deep ensemble of the deterministic matcher reaches 0.062 matching accuracy on the benchmark test set, above the competition winner; on the Bayesian matcher's assignment, the posterior flags the wrong matches that cross a cell-type boundary better than a deterministic read of the same scores (AUROC 0.68 against 0.64).

Abstract: We prove achievability below the symmetric information rate (SIR) under block maximum a posteriori (MAP) decoding for a fixed-degree MacKay-Neal (MN) socket ensemble on the quaternary insertion-deletion-substitution channel used to model DNA reads. For every integer $\ell\ge4$, the $(\ell,3,3)$ socket ensemble, with one transmitted bit fixed and an independent uniform known offset, has limiting rate $r_\star=6/\ell$ bits per stored nucleotide. If the joint SIR $J_t$ of a fixed number $t$ of reads under independent, uniformly distributed input nucleotides exceeds $r_\star$, the ensemble-average full extended-word block error is $O(1/n)$, and deterministic graph-offset sequences attain the same limiting rate and message-average error bound. This result concerns the discrete rate family $6/\ell$, not arbitrary rates approaching a given SIR. The channel permits arbitrarily many insertions and unbounded drift, with no alignment supplied to the receiver. The proof combines established MN parity and distance estimates with a strict arbitrary-candidate-list bound and a list construction that accounts for hidden output boundaries.

Abstract: The Burrows-Wheeler transform (BWT) is the core of many compressed full-text indexes and most read aligners. The fastest BWT constructions on GPUs need about $20$ bytes of device memory per character, which severely limits the texts that can be processed there. For example, on our GPU with 24 GB device memory, this is only about 1200 MiB. Prefix-free parsing (PFP) can be used to reduce the memory requirements of BWT construction algorithms in main memory. But, it uses hash tables and inverted lists which are not suited for GPUs. We show that PFP-based BWT construction can work with only sorting, scanning, and simple data-parallel procedures. This results in our GPU construction gpupfpbwt that on an NVIDIA RTX 3090, reduces the peak memory on an English text 11.9 bytes per character. There, it can processes 2108 MiB input files, which is 1.8 times larger than what the current state of the art libcubwt can handle. It is faster than or on par with libcubwt on eight of nine repetitive texts but 2.0 to 3.2 times slower on five of six non-repetitive ones, i.e., it provides a trade-off between required device memory and throughput.

Abstract: Protein annotation agents must choose suitable bioinformatics tools, interpret their outputs, and use the resulting evidence in the final answer. Most existing systems rely on frontier-scale models. We ask whether a compact model can instead learn this behavior through post-training with real tool interaction. We present ProtAgent, a two-stage pipeline that uses supervised fine-tuning (SFT) to learn the tool interface and standard on-policy REINFORCE to improve the interaction policy. Intermediate tool calls receive action rewards, and the final response receives a terminal answer reward. Step-level returns assign these rewards to the responses that produced them and to preceding actions. The episode-level baseline instead assigns one episode return to every action. On a fixed closed-loop protocol with real bioinformatics tools, ProtAgent-8B achieves 61.00% success under the annotation-coverage metric, compared with 41.60% for DeepSeek-V3.2-Thinking 671B and 60.80% for the strongest evaluated frontier system. In the three-seed comparison, mean success is 42.70% with episode-level returns and 61.40% with step-level returns under identity return normalization; the corresponding group-relative results are 36.50% and 58.20%. The metric measures recovery of reference annotations under this protocol; it does not by itself establish biological correctness or complete evidence grounding.

Abstract: The goal is to investigate whether the total human gene-score dataset and the selected human kinase-gene dataset exhibit the same number of clusters when analysed using the same set of 769 features. Two datasets were considered: the complete gene score dataset and a subset containing selected kinase genes. Two missing-data strategies were investigated. In Case I, rows containing missing gene-score values were removed. In Case II, rows with more than 20% missing values were excluded and the other missing values were replaced by the corresponding column mean. For each case, the data were scaled and principal component analysis (PCA) was performed before clustering. Four clustering approaches were considered: K-Means, Hierarchical Clustering, Fuzzy C-Means, and Graph-based Clustering. Across both datasets and both missing-data treatments, the optimal number of clusters was found to be two. In the absence of an optimal number of clusters, the quality of clustering was assessed using the Silhouette score. Relatively high Silhouette scores were obtained in all cases.

Abstract: In a seminal paper, Kratsch and Spinrad [SODA 2003 and SIAM Journal on Computing 2006] studied several important problems in structural graph theory that anticipated two trends in fine-grained complexity, both from the upper- and lower-bound side: * From the upper-bound side, for each of these problems the authors give an intermediate running time strictly between $n^ω$ and $n^3$, where $ω<2.371177$ is the matrix multiplication exponent. When optimized, this running time can be made $\tilde{\mathcal{O}}(n^{(3+ω)/2})$. By now, many more problems can be solved in this running time: directed unweighted or node-weighted APSP, a variety of matrix products such as max-min and dominance, RNA Folding and many more. For some of these, the community has also developed tools suggesting that this intermediate running time is necessary. * From the lower-bound side, the authors give lower bounds of $n^{ω-o(1)}$, conditioned on three natural hardness hypotheses on the complexity of quantified variants of triangle detection that account for different types of hardness of such problems. Despite this, it has remained open for over 20 years which of these two pieces of evidence reflects the truth for the problems that Kratsch and Spinrad study: whether the intermediate running time is indeed necessary and stronger lower bounds hold, or whether the hardness evidence of Kratsch and Spinrad is essentially tight. In this paper, we resolve this question and show that the latter is the case: all problems admit an $\tilde{\mathcal{O}}(n^ω)$ time algorithm. As an additional contribution, we make partial algorithmic progress on further problems in structural graph theory of intermediate complexity, improving over previous work [Chiu and Lu, STACS 2022 and Information and Computation 2024][Spinrad, Discrete Applied Math 2004].

Abstract: Expected-reward maximization in reinforcement learning concentrates probability mass on a few outcomes, even in tasks that require many distinct, high-quality solutions, a failure known as mode collapse. Inverse Probability Scaling (IPS) counters this in group-based policy gradients such as GRPO by dividing each reward by the outcome's probability, estimated from its frequency within the sampled group. When the outcome space is large relative to the group, outcomes rarely repeat and these estimates become uninformative. Since an outcome can be constructed along many trajectories, its probability is a sum over all of them, and computing it directly is intractable. We propose Marginal IPS (MIPS), which performs marginal evaluation by combining the sampled trajectory with a learned model of the trajectories that lead to each outcome. On Hypergrids, where the exact sampling distribution is computable, MIPS stays near the reward-distribution target as the outcome space grows relative to the rollout group, while frequency-based IPS degrades. In molecular synthesis and phylogenetic inference, MIPS sustains diverse, high-quality sampling where GRPO and frequency-based IPS collapse, showing that MIPS serves as a reliable solution to mode collapse in RL.

Abstract: Cancer diagnosis and treatment increasingly rely on multi-modal diagnostic technologies that generate high-dimensional datasets. Although these datasets are rich in pathological information, they pose considerable challenges in interpretation, dimensionality reduction, and clinical applicability. Conventional machine learning approaches often fail to extract meaningful cross-modal features, thereby limiting predictive accuracy in patient risk stratification. To address these limitations, this paper proposes an unsupervised machine learning framework that integrates feature extraction, tensor decomposition, and clustering to improve cancer risk prediction. The framework employs tensor decomposition to fuse heterogeneous genomic and clinical data, enabling the extraction of latent, biologically meaningful features that outperform unimodal feature sets. Multiple feature extraction strategies are investigated, and the resulting features are subsequently clustered using a range of clustering algorithms. Comparative evaluations based on internal and external metrics are then used to identify the most effective configurations. The framework is validated on LinkedOmics data comprising 11,158 patients across 32 cancer types. For patient risk stratification, clinically relevant attributes such as tumor purity, pathological stage, and survival outcomes serve as ground-truth indicators. Experimental results demonstrate the superiority of tSVD+GMM, which consistently outperforms other configurations across clinical benchmarks.

Abstract: Repairing scientific models from functional observations differs fundamentally from supervised prediction: feedback may certify a solution without revealing which structural correction is responsible. We study this setting for genome-scale metabolic model (GEM) repair, where multiple reaction edits can explain the same phenotypes and many apparently distinct edits correspond to the same biological mechanism. This many-to-one structure creates a hidden failure mode for conventional exploration: diversity in the output space need not translate into diversity of scientific hypotheses. We introduce QuotientPO, which collapses equivalent repairs into canonical mechanisms and optimizes exploration directly over the resulting quotient space. To make quotient exploration informative under finite rollouts, we derive a kernelized Rényi estimator that resolves graded crowding among distinct repair cores beyond coarse exact-match counts. On 2,212 held-out GEMs, QuotientPO improves Success@32 from 17.93% to 20.10% (+12.1% relative) while consistently increasing distinct successful-core discovery under the same sampling budget. These results establish quotient-space exploration as a principled approach to mechanism-level discovery under verifier-induced equivalence.

Abstract: Linkage algorithms for hierarchical clustering (HC) are a powerful and efficient framework for constructing clustering trees, yet it is often unclear which merge rule best suits a given dataset or task. In contrast, neural approaches can learn from data, but often fail to retain the efficiency and size generalization of classical algorithms. We introduce NN-linkage, a neural network (NN) model that can learn task-specific and locally dependent merge rules while retaining the recursive structure and efficient inference of classical linkage algorithms. In particular, our model is algorithmically aligned with the Lance-Williams (LW) recurrence, a parameterized framework for defining a broad, continuous family of linkage rules for agglomerative HC. Classical methods such as single linkage (SL), complete linkage (CL), and average linkage arise as discrete choices within this broader family. We show that NN-linkage is a universal approximator for continuous linkage functions, including LW recurrences, and, when paired with a transformer encoding, can also approximate globally dependent rules such as robust single-linkage. We further show that NN-linkage can exactly implement any symmetric constant-coefficient LW recurrence across all input sizes. On the empirical front, we evaluate NN-linkage in real-world applications, clock-tree routing and phylogenetic reconstruction, using both synthetic and real datasets, demonstrating its effectiveness over both classical algorithms and other neural approaches. By learning merge rules directly from target trees, NN-linkage extends efficient HC to scientific and engineering objectives not adequately captured by existing hand-designed linkage rules.

Abstract: Autoregressive transformers remain comparatively weak for protein sequence and structure generation. We study the role of target representation: amino acid tokens encode residue identities without explicit contextual semantics, while backbone coordinates require a discrete representation in our framework. We introduce two learned latent protein languages. Protein Latent Language (PLL) maps sequences to a 4,096-state contextual alphabet built on a frozen ESM-2 encoder, with one token per residue. Structure Latent Language (SLL) adapts GCP-VQVAE Lite with auxiliary sequence and confidence supervision while retaining decoding to backbone coordinates. We separately pretrain autoregressive transformer models on PLL and SLL tokens using next-token prediction, yielding PLLM and SLLM. Under matched downstream sequence training, PLLM has a fitted compute-scaling exponent of 0.038 versus 0.020 for the amino acid autoregressive model. In unconditional sequence generation, PLLM reduces the fraction of samples below a heuristic 1.5-bit residue-composition entropy threshold by 54% relative to the amino acid model across sampling temperatures. For sequence-to-structure prediction, replacing the original GCP-VQVAE Lite tokenizer with SLL reduces best validation perplexity by 34% under matched training. For long proteins, latent-token sampling is approximately 1,000 times faster than MSA-based AlphaFold2 in our measurements. In backbone generation, SLLM compares favorably with other generative models on diversity and novelty. We also observe early signs that using SLLM's internal token confidence for inference-time sampling can improve sequence-to-structure prediction quality beyond a single decoded sample. These results position learned latent protein languages as a promising substrate for autoregressive transformer scaling and inference-time sampling in protein generation.

Abstract: Protein function annotation needs to know which predictions to distrust, not only what a model predicts. We ask whether tissue-specific interaction structure carries that information. Our candidate signal is effective resistance, used previously to relieve over-squashing by rewiring. Across 24 tissue-specific interactomes it is dominated by inverse degree, and the degeneration deepens as the co-expression filtered network grows, with a Spearman correlation of -0.955. The residual departure from that limit exceeds degree-preserving null graphs in all 24 networks. Controlling for predictive entropy, degree, annotation cardinality, local structure and feature-only difficulty, the residual explains additional per-node loss in 19 of 24 held-out networks once a permutation floor is subtracted, at every depth, and the effect strengthens monotonically with depth. The increment reaches 0.37% of the variance the controls leave unexplained, 5.6 times a permutation floor, against 1.5 times when the model is retrained in a degree-preserving null world. Selective prediction improves negligibly. The signal is reproducible; degree degeneration bounds it.

Abstract: Compact regulatory DNA can free up space in vector payloads, reduce synthesis and assay burden, and expose which sequence features drive predicted activity. Yet most model-based nucleic-acid designers optimize fixed-length sequences through substitutions; they do not ask which bases of an existing functional element can be removed while retaining predicted activity. We define the task of sequence slimming as selecting an exact-length, order-preserving subsequence while retaining activity. Modeled on the design benchmark NucleoBench, we propose a quantitative evaluation for slimming that balances sequence reduction with maintaining function. Each slimmer must return both the subsequence and its source indices, which can be used to verify that the slimmer obeyed task requirements. To our knowledge, this is the first dedicated benchmark of this deletion-only problem. The coding agent Empirical Research Assistant (ERA) then searched over executable designer programs. ERA received the task prompt and a successful substitution-only designer GrAdaBeam as a starting program, and it modified the designer to produce GRADASLIM. We report held-out evaluations for five transcription-factor binding targets, comparing random, greedy, and ERA-guided slimming at 400 and 100 bp. ERA has the highest mean in 9/10 settings. Paired bootstrap intervals for ERA minus greedy are above zero in all five 400-bp settings, below zero in one 100-bp setting, and overlap zero in the remaining four.

Abstract: Single-cell transcriptomics enables profiling of cellular states at unprecedented resolution, but its high dimensionality, sparsity, and technical batch effects pose significant challenges for representation learning. Existing single-cell foundation models typically encode each cell independently or only model cells from the same batch for denoising, thereby underutilizing the rich relational information across batches and cell types to model gene expression patterns. We argue that single-cell models can benefit from more informative cell-context modeling. By comparing consistency and variation across cells, models can capture fine-grained gene-gene dependencies associated with cell states, which are essential for learning high-quality representations. Inspired by the use of multiple sequence alignment (MSA) context in protein modeling, we propose CellMSA, a single-cell representation learning framework that introduces an MSA-inspired inductive bias into transcriptomic modeling. For each target cell, CellMSA retrieves relevant cells from different batches and biologically related cell types as context, and summarizes cross-cell patterns into a context-dependent gene-pair representation. This representation is then injected into a pair-aware target-cell encoder for fine-grained representation learning. We pretrain CellMSA on a large-scale human single-cell corpus of approximately 109 million cell observations, including 65.6 million primary observations. Experiments show that our framework consistently outperforms existing methods across multiple benchmarks. Code is available at the following repository: https://github.com/PharMolix/CellMSA.

Abstract: Scientific software is increasingly required to process larger datasets while maintaining acceptable execution times. Software optimization traditionally requires substantial expertise in programming, algorithms, and numerical methods. Recent advances in large language models (LLMs) offer the possibility of automating much of this process. We investigate whether LLM-based agents can autonomously achieve substantial performance improvements in scientific software, including mature implementations that have already been extensively optimized by human developers. We tasked an LLM-based agent with optimizing software for three computational problems: t-SNE, single-sample gene set enrichment analysis (ssGSEA), and graphlet counting. Humans defined the scope, correctness criteria, and a verification mechanism, after which the agent worked autonomously, in some cases for several hours. Code maintainers reviewed each resulting implementation and verified its correctness. The optimized implementations were faster in all tested configurations, by up to two orders of magnitude over the fastest existing tools. The improvements included low-level code optimizations, mathematical reformulations, and an entirely new algorithm for graphlet counting. Software optimization can increasingly be delegated to autonomous agents, with the human role shifting from implementing optimizations to deciding which software to optimize, defining objectives, providing verification mechanisms, and ensuring the correctness of the final software. For well-scoped, verifiable problems, we argue that manual software optimization may be a thing of the past.

Abstract: Predicting Drug-Target Interactions~(DTIs) is a central task in computational drug discovery, with direct applications in virtual screening, drug repurposing, and therapeutic candidate prioritization. Although recent deep learning methods have improved DTI prediction, many sequence-based models still process drugs and proteins independently and only combine their representations at a late prediction stage. This limits their ability to explicitly model cross-molecular dependencies between chemical substructures and protein sequence regions. In this paper, we propose a sequence-only DTI prediction architecture that combines two pre-trained language models, ChemBERTa for drug SMILES strings and ESM-2 for protein amino acid sequences, with a hierarchical interaction module. The proposed model first extracts contextual representations using pre-trained encoders, then applies 1D convolutional layers to condense local sequence patterns, followed by a sequential bidirectional cross-attention mechanism inspired by the induced-fit view of molecular recognition. Finally, attention-based pooling constructs fixed-size interaction-aware vectors for binary prediction. Experiments on BIOSNAP, Davis, and BindingDB show that the proposed model achieves the best performance on BIOSNAP, matches the best AUROC on Davis, and remains competitive on BindingDB while using only 25.2 million trainable parameters. Ablation results confirm the contribution of both the CNN and cross-attention modules, and cold-start experiments indicate promising generalization to unseen proteins and drugs.

Abstract: Understanding how drugs interact with protein targets is fundamental to drug discovery, drug repurposing and the early identification of promising therapeutic candidates before costly experimental testing. Sequence-based DTI models face three practical limitations: labelled interactions are scarce and unevenly distributed, large pretrained chemical and protein encoders are expensive to fine-tune end-to-end, and independently encoded sequences do not capture pair-specific dependencies. We present BERT4DTI, which encodes SMILES strings with ChemBERTa and amino-acid sequences with ProtBERT, applies bidirectional mutual attention between token-level representations, and classifies the resulting interaction features using convolutional layers and a multilayer perceptron. To reduce trainable size, ProtBERT is truncated to 18 retained layers and only the last two layers of each encoder are fine-tuned. On BIOSNAP, DAVIS and BindingDB, BERT4DTI is competitive, achieving the best ROC-AUC and PR-AUC on BIOSNAP and the highest sensitivity on all three benchmarks. An ablation on DAVIS shows that mutual attention improves PR-AUC and specificity. With 125M trainable parameters compared with 353M for full BERT fine-tuning, BERT4DTI provides a favourable performance-parameter trade-off for sequence-based DTI screening, while leaving runtime profiling, calibration and leakage-audited validation for future work.

Abstract: The goal of personalized immunogenomics is to recover an individual's germline immunoglobulin gene segments from 'repertoire sequences' altered by trimming, extension, and mutation. In this work, we study the fundamental trace complexity (or the number of samples/traces required for accurate reconstruction) of D-gene reconstruction under three biologically motivated trace-generation models introduced by Bhardwaj et al. (2021) and develop practical algorithms for reconstruction problems left open in the original work. First, for the TrimSuffixAndExtend model, we establish the optimal trace complexity to be {Θ(n)}, and develop a low-complexity Prefix-Filtered Mode (PFM) decoder that achieves this scaling. Second, for the closely related two-sided TrimAndExtend model, we show that the optimal trace complexity is instead {Θ(n^2)}, and is achieved by the simple Bit-Wise Mode (BWM) decoder. Third, for the SuffixExtend-t(TrimSuffix) model, we establish polynomially separated lower and upper bounds on trace complexity. Our results follow from information-theoretic lower bounds coupled with tight analyses of the proposed reconstruction algorithms.

Abstract: Approximate pattern matching (APM) on highly repetitive texts is a central task in bioinformatics. Bi-directional r-indexes support left- and right-extension of a pattern and thereby accelerate APM algorithms based on search schemes, but existing variants -- br-index and b-move -- suffer from two bottlenecks: $O(σ)$ character-predecessor/-successor queries on the run-length-encoded BWT per extension, and predecessor queries on sparse bit vectors to maintain a value in the suffix array interval and to access the PLCP array while locating. We present Move-rb, a bi-directional r-index built on the optimized r-index Move-r. Although Move-rb is 2x larger than br-index, it is up to 24% smaller than b-move, answers APM queries 1-4 orders of magnitude faster than br-index, 1.9-10x faster than b-move and up to 5.5x faster than the state-of-the-art bi-directional (uncompressed) FM-index columba, which is 36-42x larger. Memory usage (including index size) during APM locate queries is reduced by 1.5x (up to 6.5x) for Hamming distance and 2.5x (up to 7.4x) for edit distance. Move-rb can be constructed 3-14x faster while using 19-141x less memory than br-index, b-move and columba. A variant using a relative Lempel-Ziv-encoded suffix array locates up to 10x faster while being 1.2-2.4x larger. We achieve these speedups by optimizing index operations and search scheme APM algorithms: Without a direction switch, Move-rb computes an all-$k$-character extension in output-optimal $O(k)$ time and a single-character extension in the same time. Augmenting Move-rb with $O((r+\overleftarrow{r})\log(n/r_{\min})\logσ)$ bits reduces a single-character extension to $O(\logσ)$ time, where $r$ and $\overleftarrow{r}$ are the numbers of runs in the BWT of the text and its reverse, and $r_{\min}=\min(r,\overleftarrow{r})$. A direction switch incurs only $O(\log\log_ω(n/r_{\min}))$ additional time.

Abstract: Motivation: The Multiple Longest Common Subsequence (MLCS) problem for an arbitrary number of sequences is an NP-hard problem in sequence analysis. Existing exact algorithms based on dynamic programming or MLCS-DAG pruning rapidly exhaust memory as sequence lengths and set sizes grow, while heuristic and hyper-heuristic approaches compromise solution quality and require heavy parameter tuning, respectively. Results: This paper presents the ARP heuristic that, for a given primary sequence, performs three key actions: (i) Adding-$Δ$s, which incrementally builds a solution by adding subsequences, called $Δ$s, of the primary sequence to an initial solution; (ii) Replacing-Subsequences, which enhances the diversity of the solution set via targeted replacement of subsequences common to all solutions; and (iii) Prioritizing groups of characters from the primary sequence that are likely to appear together in longer common subsequences. ARP allows the user to configure the degree of Replacements and Prioritization actions for carrying out quality-runtime tradeoff. Empirical evaluations on both synthetic and biological sequence sets demonstrate that ARP's fastest configuration AOnly finds significantly longer common subsequences than the BNMAS classical heuristic and its aggressive configuration attains solution quality comparable to state-of-the-art hyper-heuristic (UB-HH) while running 1.1$\times$-1.7$\times$ faster.

Abstract: Functional annotation of newly sequenced proteins remains a bottleneck in molecular biology: the number of sequences in public repositories grows far faster than the capacity for manual curation. Most computational approaches consider annotation as multi-label classification over a fixed ontology, which constrains predictions to a predefined label set. In this work we study the the protein annotation as a sequence-to-text generation problem. We fine-tune the 3B-parameter Ministral 3 base model with QLoRA (4-bit NF4 quantization with low-rank adapters) on sequence annotation pairs. We assess predictions with an LLM-as-expert protocol: a GPT model prompted as a senior molecular-biology curator scores organism identification as binary and function annotation quality. We conclude that QLoRA-fine-tuned compact LLMs can generate curator-style annotations with genuine biological value for a substantial subset of proteins. We also discuss future directions in data quality, model scaling, and evidence grounding that are needed to make the approach sufficiently reliable for practical use.

Abstract: Antibody lead optimization calls for a small, bounded set of edits to an existing candidate: substitutions, but also insertions and deletions. Edit-based generative models are the only ones that allocate such an edit budget without fixing the edit positions, the edit count, or the output length in advance. However, the existing approaches Edit Flows and EvoFlows did not release code or complete training specifications. Here, we show that both methods follow the same underlying process -- edits firing one at a time, at learned rates, in continuous time -- the pure-jump case of generator matching over finite sequences. With EditJumps we introduce the first open implementation of this framework, with a single generalist antibody editor trained on 1.66M Observed Antibody Space homolog pairs to propose homolog-like variants of a seed sequence, editing unseen leads zero-shot, without the per-family retraining original approaches require. Replicating this system from scratch exposes why open code is essential for generative biology: reconciling published edit distributions required reverse-engineering an undocumented rate-scaling hyperparameter that dictates realized mutation counts. Moreover, we show that published evaluation metrics are highly sensitive to reference sample size, frequently flipping method rankings. We release our full codebase, automated test suite, and configurations at: https://github.com/VisiumCH/editjumps

Abstract: Reinforcement learning over a frozen reasoner has become a common recipe for teaching a policy which external tools to invoke. We show that this recipe becomes structurally mismatched in specialist scientific settings where the complete tool-subset space is enumerable. There, a small set of recurring computational capabilities covers the domain, so the space of tool subsets is combinatorial yet small enough to enumerate, and GRPO still estimates an action expectation from a handful of sampled rollouts. Worse, the approximation degrades as training succeeds: as the policy concentrates on preferred subsets it resamples them, sampled rewards collide, and the group-normalized advantage vanishes. On genomic reasoning the fraction of questions yielding no reward signal rises from 0.2% under a uniform reference policy to 20.8% after GRPO training. As a remedy, we introduce FGPO (Full-Group Policy Optimization), which (1) scores every tool subset and optimizes the exact action expectation, so each update sees the complete action space, and (2) precomputes the reward of each question--subset pair into an exhaustive table, removing frozen-reasoner calls from the training loop entirely. Across five frozen reasoners and three genomic benchmarks, FGPO outperforms GRPO in all 15 settings by 6.75 points on average and up to 14.20, while a standard on-demand GRPO schedule would require 2.4 times as many frozen-reasoner reward evaluations and, on GenomeQA, FGPO cuts invoked tools per question from 2.36 to 1.40.

Abstract: Protein-protein interaction (PPI) databases do not faithfully reflect biological realities. Instead, they are influenced by study and technical biases that distort certain protein and interaction attributes. Machine learning models can exploit these as learning shortcuts if the negative dataset is not constructed with care. So far, the shortcuts introduced during PPI dataset construction have only been examined in isolation. Here, we systematically characterize both reported and, to our knowledge, previously unreported biases in PPI datasets that lead machine learning models to learn shortcuts instead of biological signal. We analyze HIPPIE, IntAct, and STRING, dedicated PPI databases, as well as two datasets derived from 3D-structural information in the Protein Data Bank (PDB). We show that random data splitting introduces strong topological shortcuts. When train-test protein overlap is removed, the resulting datasets still retain usable shortcuts stemming from self-interactions, taxonomic identity, and functional relatedness, whose prevalence interestingly depends on the data source. We further show that sampling negatives from a set of high-confidence non-interactors, an intuitively appealing choice, can amplify the shortcut stemming from functional relatedness. To detect and mitigate these biases, we provide an open Nextflow pipeline that combines similarity-aware, data-loss-minimizing dataset splitting with bias-minimizing negative sampling, both formulated as integer linear programs. Its key concept of quantifying biases to minimize them through optimization-based negative sampling can, in principle, be extended to any machine learning problem where the pool of negative candidates is much larger than the positives and is thus of interest also beyond PPI prediction.