Machine Learning in Bioinformatics

Cataloging cysteines in ECOD domains using a protein language model

The paper develops TriCyP, a model to predict cysteine states in protein structures, achieving high accuracy and identifying new biological patterns and protein families.

Protein solubility depends on centrifugation: Aiki-Sol, a per-regime predictor for E. coli

The paper introduces Aiki-Sol, a predictor for E. coli protein solubility that accounts for centrifugation regimes, improving performance by treating regime as a feature rather than noise.

Structure tokens sharpen the feature vocabulary of protein language models

The paper shows that structure tokens in ESM-3 enhance the shared biological vocabulary of protein language models without creating a new feature set.

Task-Specialized Protein Language Models Decode the Sequence Grammar of Post-Translational Modification Sites

The paper fine-tunes a protein language model for PTM-site prediction and shows how model size, annotation depth, and modification chemistry affect PTM recognition.

BindPred: A Framework for Predicting Protein-Protein Binding Affinity from Language Model Embeddings

BindPred predicts protein-protein binding affinity from amino acid sequences using language model embeddings and gradient boosting trees, achieving high correlation on a benchmark dataset.

Investigation of Protein Melting Temperature Prediction with Cross-Method Validation on Biophysical Data

The paper evaluates deep learning models for predicting protein melting temperatures using proteomics and biophysics datasets, finding that fine-tuned embeddings improve performance on diverse validation sets.

PTM-dCN: Latent Space Control for Post-translational Modification-aware Protein Design

This paper introduces a deep learning framework for designing proteins with site-specific post-translational modifications using latent diffusion models and a PTM-aware language model.

BERT-T6: Towards High-accuracy T6SS Bacterial Toxin Identification Using Protein Language Model

The paper develops BERT-T6, a model using transfer learning on ProtBert embeddings to predict T6SEs with high accuracy and performance metrics.

ProtSpace: Protein Universe in Your Browser

ProtSpace visualizes protein language model embeddings interactively, enabling hypothesis generation across biological scales without installation.

GENERator-v2: Reconciling Coarse Tokenization with Single-Nucleotide Resolution in Genomic Language Modeling

This paper introduces improved genomic language models that balance long-range context with single-nucleotide resolution, achieving better performance and efficiency than prior methods.

An ensemble method for predicting and designing of druggable proteins.

The paper develops an ensemble method combining machine learning and motif analysis to predict druggable proteins, achieving high accuracy on an independent validation dataset.

Diagnosing protein sequence search in the era of language models

This study evaluates the reliability of protein sequence search using language models and finds that classical alignment methods are more robust than PLM-based approaches.

Protein Function Prediction with Pretrained ProtT5 Embeddings and Gradient Boosting

This study evaluates using ProtT5 embeddings with gradient boosting for Gene Ontology prediction, finding gradient boosting provides the best balance between performance and stability.

PROBind: A Web Server for Prediction, Analysis and Visualization of Protein-Protein and Protein-Nucleic Acid Binding Residues

PROBind predicts, analyzes, and visualizes protein-protein and protein-nucleic acid binding residues using sequence and structure data, integrating multiple predictors and providing interactive tools.

GenNA: Conditional generation of nucleotide sequences guided by natural-language annotations

GenNA generates nucleotide sequences using natural-language annotations, showing improved performance in semantic and species context alignment without explicit biological rule supervision.

Fine-Tuning Protein Language Models Enhances the Identification and Interpretation of the Transcription Factors

This paper introduces a two-layer framework using protein language models to identify transcription factors and predict TFs with methylated DNA binding preferences, achieving better performance than existing methods.

Aiki-XP: leakage-controlled multimodal prediction of within-species relative protein expression at pan-bacterial scale

Aiki-XP predicts within-species relative protein expression across bacteria using multimodal data, outperforming single-omics models in leakage-controlled settings.

Back to BERT in 2026: ModernGENA as a Strong, Efficient Baseline for DNA Foundation Models

This paper adapts a modernized BERT-style model to genomic sequence modeling, achieving strong efficiency and performance, and releases it as an open baseline.

BioAutoML-FAST: an automated machine-learning platform for reusable and benchmarked biological sequence models

BioAutoML-FAST provides an automated ML platform for biological sequence analysis, offering reusable models and benchmarking across genomic, transcriptomic, and proteomic tasks.

PEACE: Prototype-aware Effector Analysis via Contrastive Embeddings

PEACE improves effector protein prediction by using contrastive embeddings and prototype-aware training, outperforming EffectorP 3.0 on fungal and fungi+oomycete datasets.

Hybrid Gated Fusion: A Multimodal Deep Learning Framework for Protein Function Annotation

Hybrid Gated Fusion improves protein function annotation by integrating sequence, structure, text, and interaction data using bilinear gating and auxiliary supervision, achieving state-of-the-art results on CAFA3.

Machine learning for optimal growth temperature prediction of prokaryotes using amino acid descriptors

This study improves OGT prediction for prokaryotes using machine learning with amino acid descriptors, achieving an R2 of 0.89.

Structure-aware geometric graph learning for modeling protease-substrate specificity at scale

OmniCleave models protease-substrate specificity using a geometric graph framework, outperforming existing methods and identifying novel substrates.

StarFunc: fusing template-based and deep learning approaches for accurate protein function prediction

StarFunc combines deep learning with template information from multiple sources to improve protein function prediction, outperforming existing methods in benchmark tests.

From nucleotides to semantics: genomic representation learning via joint-embedding predictive architecture

This paper introduces GenoJEPA, a genomic representation learning framework that improves efficiency and performance in downstream tasks by focusing on semantic alignment rather than nucleotide-level reconstruction.

Protein Language Models Outperform BLAST for Evolutionarily Distant Enzymes: A Systematic Benchmark of EC Number Prediction

The paper benchmarks protein language models for EC number prediction, finding MLP classifiers outperform BLAST and PLMs show significant gains for distant eukaryotes.

Quantifying uncertainty in Protein Representations Across Models and Task

This study evaluates protein embeddings' reliability by introducing a model-agnostic scoring framework to quantify their ability to capture biologically relevant information.

eSIG-Net: Accurate prediction of single-mutation induced perturbations on protein interactions using a language model

eSIG-Net predicts how single mutations affect protein interactions using a sequence-based interaction language model, outperforming existing methods.

Reliable prediction of short linear motifs in the human proteome

SLiMMine predicts short linear motifs in the human proteome with high reliability, reducing false positives and discovering novel motifs.

AINN-P1: A Compact Sequence-Only Protein Language Model Achieves Competitive Fitness Prediction on ProteinGym