Open research questions in Machine Learning in Bioinformatics

What the literature leaves open

• The capabilities of LLMs in bioinformatics are unknown. There is a lack of evaluation of LLMs on basic bioinformatics tasks that require multi-step reasoning and code writing. The performance of LLMs compared to humans on bioinformatics tasks is unknown.

  Assessment of LLMs on basic bioinformatics tasks requiring multi-step reasoning and code writing · 2026 · DOI

• The need to incorporate positional information of k-mers. The comparison with existing state-of-the-art methods.

  PhageCGRNet: Integrating Chaos Game Representation of Genomes with Convolutional Neural Network for accurate phage host classification prediction · 2026 · DOI

• Based on the limitations discussed above, we outline three directions for future work: (i) multimodal dataset construction, (ii) meta-learning to address data scarcity and class imbalance, and (iii) generalizability assessment and uncertainty quantification.

  Machine learning for the prediction of gram-negative bacterial secreted effectors: advances and challenges · 2026 · DOI

• The performance of UNKAI is limited for inputs that lack similar samples in the training set. The distribution of functional categories in the training set exhibits some bias. The model's predictive capabilities for entirely unseen EC subclasses are lower than the overall performance.

  UNKAI: A protein functional identity prediction model based on ESM-C latent representations and the attention mechanism · 2026 · DOI

• To improve the performance of UNKAI for inputs that lack similar samples in the training set. To apply UNKAI to various fields, such as biochemistry, biophysics, and pharmacology. To develop new methods that can leverage pLMs and attention mechanisms for protein functional identity prediction.

  UNKAI: A protein functional identity prediction model based on ESM-C latent representations and the attention mechanism · 2026 · DOI

• We do not claim parity with specialist remote-homology tools; published numbers for ESM-2, CATHe and PLMSearch on differently constructed splits reach 65--75%, and closing this gap is discussed as an open problem.

  OmniGene-4: A Unified Bio-Language MoE Model with Router-Level Interpretability · 2026 · DOI

• While we have established that Lutein inhibits Th1 cell function by blocking glucose uptake, the precise upstream regulatory mechanisms remain unknown.

  Few-shot learning-driven discovery of Lutein suppresses Th1-mediated inflammation via glucose metabolism · 2026 · DOI

• The complexity of data preprocessing pipelines and the lack of unified implementations make methods and results difficult to reproduce and compare. The need for a unified framework to analyze different approaches for molecule retrieval from MS/MS spectra.

  MSAlign: Aligning Molecule and Mass Spectra Foundation Models for Metabolite Identification · 2026

• The scarcity of clinical data limits the prediction of patient-level drug response. Biological differences introduce domain shifts that hinder direct translation to patient tumors.

  Deep learning for predicting patient drug response by transferring gene-level and cell-level knowledge to tumors · 2026 · DOI

• Prior datasets have limitations, such as limited annotation depth and scope. There is a need for a comprehensive and enriched resource for peptide hormone research.

  HORDB 2.0: a comprehensive dataset of peptide hormones · 2026 · DOI

• While protein generation has broadly followed trends in NLP, two directions remain underexplored: alignment methods that optimize model behavior toward design objectives, and prompting-based control at inference time without fine-tuning.

  ProtGPT3: an Open-source family of Promptable and Aligned Protein Language Models · 2026 · DOI

• The dodecin signal is therefore a partial-subdomain match limited by the size of the dodecin chain rather than a full-topology alternative. The heatmap is sparse, with most cells exactly zero, and the non-zero signal concentrates on the top rows of the plot in the same four cases the prose above describes, namely P.

  Tokens, Topologies, Taxa: Towards Declarative Biology and Bioengineering · 2026 · DOI

• The study suggests the need for further research in diverse populations. The study proposes the application of augmentation strategies to enhance generalizability. The study recommends the development of new computational predictors for other diseases with a genetic component.

  Comparing GATs and transformers for predicting functional COMT sequences in temporomandibular joint pain · 2026 · DOI

• The study highlights the need for further research in diverse populations. The study acknowledges the limitations of prior studies, including the small number of studies.

  Comparing GATs and transformers for predicting functional COMT sequences in temporomandibular joint pain · 2026 · DOI

• Although the OGTs for most bacteria remain unknown, the increasing availability of genomes from uncultivated and cultivated taxa has made it advantageous to build genomic, cultivation-independent models to infer OGT.

  Predicting optimal growth temperatures of bacteria using learned structural information from a single protein · 2026 · DOI

• The challenge of reconstructing genome-scale TRN is a highly challenging systems engineering task. The core difficulty lies in the heavy reliance on massive experimental data. The challenge is particularly significant for non-model organisms lacking extensive transcriptomic data.

  DDTRN: Predicting Bacterial Transcriptional Regulatory Networks Based on Gene Sequences using Dual Descriptor · 2026 · DOI

• Accurate computational reconstruction of bacterial transcriptional regulatory networks remains a challenge. The challenge is particularly significant for non-model organisms lacking extensive transcriptomic data.

  DDTRN: Predicting Bacterial Transcriptional Regulatory Networks Based on Gene Sequences using Dual Descriptor · 2026 · DOI

• To explore the application of PhageCGRNet in other fields such as microbiology and genomics.

  PhageCGRNet: Integrating Chaos Game Representation of Genomes with Convolutional Neural Network for accurate phage host classification prediction · 2026 · DOI

• Downstream applications of the dataset include comparative genomic and proteomic studies. The dataset can be used for training machine learning models to predict hormone activity, stability, or other bioactive properties.

  HORDB 2.0: a comprehensive dataset of peptide hormones · 2026 · DOI

• For fitness landscape prediction, accuracy may decrease in regions of high epistatic complexity with sparse training data; for antibody–antigen prediction, performance may be limited for underrepresented antibody classes and does not account for post-translational modifications or glycan shielding.

  Predicting the Evolutionary and Functional Landscapes of Viruses with a Unified Nucleotide-Protein Language Model: LucaVirus · 2026 · DOI

• Investigating the performance of LLMs on more complex bioinformatics tasks, - Evaluating the effectiveness of LLMs in real-world bioinformatics applications, - Exploring the potential of LLMs to accelerate biological research, - Investigating the use of RAG models in different biological…

  Assessment of LLMs on basic bioinformatics tasks requiring multi-step reasoning and code writing · 2026 · DOI

• Exploring the application of the approach to other bacteria and domains. Evaluating the performance of the approach using larger datasets and more complex TRNs.

  Fine-tuned GPT-based foundation models effectively reconstruct bacterial transcriptional regulatory networks from literature · 2026 · DOI

• The role of MERCs in carcinogenesis still remains unknown - The current interface should be interpreted as an informal pilot assessment rather than a formal usability…

  Cell Structure Segmentation in TEM Images of Murine Skin Melanoma Cells by Deep Learning Model · 2026 · DOI

• Search for new tools facilitating the study of MERCs in tumor cells - Evaluation of the ability of the network models to differentiate cellular structures after pre-training on external…

  Cell Structure Segmentation in TEM Images of Murine Skin Melanoma Cells by Deep Learning Model · 2026 · DOI

• Further analysis of the sources of information supporting strong benchmark performance. Development of more effective and interpretable models that take into account the contributions of different information sources.

  Performance Attribution in pLM-Based Biological Relation Prediction · 2026 · DOI

Most-cited papers in Machine Learning in Bioinformatics

• Evolutionary-scale prediction of atomic-level protein structure with a language model · Science · 2023 · 4,885 citations
• A novel genetic system to detect protein–protein interactions · Nature · 1989 · 4,762 citations
• Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences · Proceedings of the National Academy of Sciences · 2021 · 3,268 citations
• Improved protein structure prediction using potentials from deep learning · Nature · 2020 · 3,127 citations
• Highly accurate protein structure prediction for the human proteome · Nature · 2021 · 2,936 citations
• SignalP 6.0 predicts all five types of signal peptides using protein language models · Nature Biotechnology · 2022 · 2,929 citations
• Large language models generate functional protein sequences across diverse families · Nature Biotechnology · 2023 · 1,095 citations
• Learning the protein language: Evolution, structure, and function · Cell Systems · 2021 · 524 citations
• iPHoP: An integrated machine learning framework to maximize host prediction for metagenome-derived viruses of archaea and bacteria · PLoS Biology · 2023 · 492 citations
• Applying and improving AlphaFold at CASP14 · Proteins Structure Function and Bioinformatics · 2021 · 428 citations

Most recent work

• A multispecies foundation model for genome function prediction, gene annotation and sequence design · bioRxiv · 2026
• Protein and genomic language models uncover the unexplored diversity of bacterial immunity · Science · 2026
• Deep capsule neural network for identifying anticancer peptides using sequence to image transformation-based local embedded features · BMC Biology · 2026
• Whole-genome prediction of bacterial pathogenic capacity on novel bacteria using protein language models with PathogenFinder2 · Bioinformatics · 2026
• The Dayhoff Atlas: scaling sequence diversity for improved protein generation · bioRxiv · 2026
• Clustering the protein universe of life using DIAMOND DeepClust · Nature Methods · 2026
• Dual-Modal Fusion Based on Peptide Sequence and Molecular Graph with Bidirectional Cross-Attention for Precise Peptide Toxicity Prediction · Journal of Chemical Information and Modeling · 2026
• Learning heterogeneous biological interactions via meta-relation-guided dual-channel graph transformer for circular ribonucleic acid function prediction · Knowledge-Based Systems · 2026
• Machine learning for enzyme catalytic activity: current progress and future horizons · Briefings in Bioinformatics · 2026
• The DNA dialect: a comprehensive guide to pretrained genomic language models · Molecular Systems Biology · 2026

Related topics in Biochemistry, Genetics and Molecular Biology