computational biology

Machine learning (ML) applications in upstream biopharmaceutical processes are hindered by scarce, costly-to-acquire, and mechanistically complex process data. Method: This paper systematically reviews ML methodologies tailored for few-shot learning scenarios in bioprocessing. We propose the first taxonomy of ML methods specifically designed for bioprocess small-data regimes, organizing techniques along three unified dimensions: data augmentation, transfer learning, and physics-guided modeling—including meta-learning, Bayesian optimization, physics-informed neural networks (PINNs), few-shot transfer learning, and synthetic data generation. Contribution/Results: We identify and categorize 12 applicable methods, empirically evaluating their performance on critical tasks such as cell culture titer prediction and key process parameter optimization. The study reveals significant gaps in interpretability, cross-process generalizability, and experimental validation. Our analysis delivers a theoretically grounded, industrially actionable framework for method selection in biomanufacturing.

This paper addresses the challenges of data scarcity, high computational cost, and difficulty in cross-omics integration that hinder large language model (LLM) adoption in bioinformatics. To this end, we propose the first unified LLM framework tailored to multimodal biomolecular data—including DNA, RNA, proteins, and single-cell transcriptomes—integrating Transformer architectures, prompt engineering, parameter-efficient fine-tuning, and multi-task pretraining. We further introduce a novel evaluation paradigm specifically designed for biological data characteristics, benchmarking over 200 LLM-driven methods across four core tasks: genomic sequence modeling, RNA structure prediction, protein function inference, and single-cell transcriptomic analysis. Our results delineate performance boundaries and clarify translational pathways to clinical applications. Collectively, this work provides both theoretical foundations and practical guidelines for developing scalable, interpretable, and cross-omics–cooperative biomedical AI foundation models.

In biological research, the fragmentation between statistical analysis and machine learning tools, coupled with high usability barriers for non-programming users, impedes efficient and rigorous data-driven discovery. Method: We propose BioAutoML, a modular, biology-oriented automated analysis platform integrating classical statistical methods (e.g., t-tests, ANOVA, Pearson correlation) with interpretable machine learning (e.g., Random Forest classification). It supports automated data preprocessing, categorical encoding, feature importance assessment, and data-aware dynamic model configuration. Crucially, it introduces the first unified statistical–machine learning workflow, bridging methodological gaps via automated hyperparameter optimization. Contribution/Results: Evaluated on multiple chemomics datasets, BioAutoML achieves significantly higher classification accuracy than baseline approaches while preserving statistical validity. It enables domain scientists without programming expertise to perform end-to-end, interpretable, and statistically sound modeling—substantially accelerating biological insight generation.

This study addresses the challenge of effectively integrating heterogeneous proteomic data—specifically, whole-sample mass spectrometry (MS) and multiplexed protein array profiles—in pancreatic cancer research. To overcome the limitations of conventional approaches that naively concatenate multi-source features, the authors propose a novel model fusion framework that explicitly models and leverages the heterogeneity between data sources through a tailored integration strategy. This approach synergistically exploits the complementary strengths of each modality rather than treating them as homogeneous inputs. Experimental results demonstrate that the proposed method significantly outperforms both single-modality models and standard fusion baselines in pancreatic cancer classification, yielding substantial improvements in diagnostic accuracy. The work thus offers a principled and effective paradigm for integrating heterogeneous multi-omics data in biomedical applications.

Existing public cancer multi-omics databases (e.g., TCGA, LinkedOmics) lack machine learning–oriented standardized preprocessing and unified programmatic interfaces, hindering model development and reproducibility. To address this, we propose MLOmics—the first plug-and-play multi-omics machine learning benchmark for cancer. It integrates data from 8,314 patients across 32 cancer types sourced primarily from TCGA, spanning genomic, transcriptomic, methylomic, and proteomic modalities. MLOmics provides standardized data curation, cross-platform feature alignment, stratified sampling, and integration with the BioLink knowledge graph. The open-source benchmark includes a complete end-to-end data pipeline, performance reports for 12 baseline models, and illustrative cross-omics fusion analysis cases. By unifying data access, preprocessing, and evaluation protocols, MLOmics significantly improves model reproducibility, cross-cancer generalizability, and biological interpretability—thereby bridging the critical gap between multi-omics resources and ML-driven cancer research.

Addressing the challenge of simultaneously achieving high prediction accuracy and biologically interpretable biomarker identification in high-dimensional genomic binary classification, this paper proposes the first data-driven logistic regression (LR) ensemble framework that jointly ensures statistical interpretability and strong predictive performance. The method integrates L₁/L₂ regularization with ensemble learning to directly learn a small set of highly accurate and interpretable LR models via global optimization. It provides the first rigorous derivation of asymptotic statistical properties for such regularized ensemble LR estimators. Additionally, we develop a gene importance ranking tool based on resampling stability to enhance biological interpretability. Evaluated on real-world datasets—including cancer, multiple sclerosis, and psoriasis—the framework achieves significant improvements in classification accuracy and successfully identifies several critical disease-associated genes—previously missed by competing methods—that are independently validated in the biomedical literature.

Existing gene sequence alignment methods lack systematic, cross-platform evaluation. Method: We introduce the first benchmark platform specifically designed for multi-platform sequencing data (Illumina, Oxford Nanopore Technology, and PacBio), systematically evaluating 11 state-of-the-art aligners—including exact, heuristic, and learning-enhanced algorithms. Contribution/Results: Our end-to-end empirical analysis quantifies, for the first time, the high sensitivity of alignment performance to both sequencing data quality and hyperparameter configurations. We propose a standardized four-dimensional evaluation framework assessing accuracy, speed, memory footprint, and noise robustness. Results reveal widespread deficiencies in robustness and resource efficiency across current tools. To support reproducibility and methodological advancement, we open-source a fully documented, end-to-end benchmarking pipeline on GitHub—providing an evidence-based foundation for algorithm selection, comparative analysis, and future alignment method development.

In clinical proteogenomics, converting raw multi-omics data into reliable, novel biological hypotheses remains a major challenge due to the lack of automated, interpretable frameworks. Method: We propose PROTEUS—the first fully automated hypothesis generation framework that uniformly models the scientific discovery process as an evolvable, interpretable research process graph. It integrates large language models, modular workflow simulation, graph neural network–based representation learning, and an automatic open-scoring mechanism to enable end-to-end analysis of heterogeneous high-throughput data. Contribution/Results: PROTEUS unifies exploratory analysis, statistical testing, and iterative hypothesis generation within a single graph structure, supporting open-science–driven autonomous discovery. Evaluated on 10 public clinical multi-omics datasets, it generated 360 hypotheses; external validation and automated assessment demonstrated significant improvement in the reliability–novelty trade-off. This advances general-purpose AI toward domain-specialized scientific discovery systems.

Euclidean space struggles to capture the nonlinear hierarchical structure inherent in biological sequences, limiting performance in sequence classification and similarity measurement. To address this, we propose a hyperbolic representation framework for genomic sequences based on the Poincaré ball model. Our method employs a learnable hypersurface feature mapping to embed discrete sequences into continuous hyperbolic space, preserving their intrinsic tree-like or hierarchical topology while achieving substantial dimensionality reduction. We further introduce a hyperbolic inner-product-based kernel matrix to enable efficient and geometrically consistent pairwise sequence similarity modeling. Experiments across multiple benchmark datasets demonstrate that our approach achieves an average 5.2% improvement in classification accuracy over Euclidean baselines and outperforms existing hyperbolic embedding methods in capturing biologically meaningful sequence correlations. This work establishes a theoretically grounded and practically effective paradigm for biological sequence analysis.

Current biomolecular sequence design methods lack unified, reproducible evaluation standards, hindering fair and rigorous performance comparison. To address this, we introduce BioSeqEval—a modular, open-source Python evaluation library that systematically integrates three model-agnostic metric categories: sequence-based, embedding-based, and property-based—representing the first such comprehensive framework. It supports one-shot and iterative design evaluation across diverse sequence modalities, including small molecules, DNA, RNA, peptides, and proteins. The library incorporates state-of-the-art pretrained embedding models, machine learning–based property predictors, efficient sequence alignment tools, and interactive visualization modules for diagnostic analysis. Empirical evaluation demonstrates that BioSeqEval significantly enhances evaluation standardization, cross-method comparability, and methodological transparency. It exhibits strong flexibility and robustness across multiple benchmark design tasks, enabling reproducible, interpretable, and scalable assessment of generative sequence models.

To address the bottleneck of supervised learning—its reliance on large-scale labeled samples—in phenotypic prediction from gene expression data, this study proposes a novel self-supervised representation learning paradigm. We systematically adapt three state-of-the-art self-supervised learning strategies—contrastive learning, masked reconstruction, and generative modeling—to bulk RNA-Seq data for the first time, conducting pretraining and downstream fine-tuning across multiple public datasets. Experimental results demonstrate that our approach significantly outperforms fully supervised baselines in phenotypic prediction accuracy; notably, it achieves comparable performance using only 10% of labeled data. Each method exhibits distinct strengths: contrastive learning delivers superior generalizability, masked reconstruction shows robustness to sparse signals, and generative modeling better supports multi-omics integration. This work establishes a reproducible methodological framework and practical guidelines for label-efficient analysis of genomic data.

This study investigates the evolutionary dynamics and structural characteristics of open-source machine learning models during downstream fine-tuning. Leveraging metadata and model cards from 1.86 million models on Hugging Face, it pioneers the application of evolutionary biology frameworks to AI model ecosystems—constructing fine-grained model phylogenies and integrating network analysis, genetic similarity metrics, and textual feature extraction to systematically characterize inheritance, mutation, and diffusion patterns. Key findings include: (1) a trend toward more permissive licensing; (2) significant degradation in multilingual support, with increasing English dominance; and (3) growing template-driven homogenization of model documentation. The work further identifies “family resemblance” as a structural invariant and reveals directed, rapid mutational bursts—demonstrating that AI model evolution follows quantifiable, predictable systemic drift. These insights establish a novel paradigm for model governance, reproducibility assessment, and ecosystem health monitoring.