Machine Learning in Bioinformatics

AltereGO: Prediction of Protein Function by Multimodal Fusion of Protein Language Models.

AltereGO fuses protein language models and structure representations to predict molecular functions, achieving better performance than individual models and some existing methods.

HERCULES: a deep-learning framework for RNA-binding region localization and mutation-effect prediction

HERCULES predicts RNA-binding regions and mutation effects using a deep-learning framework combining language models and physicochemical descriptors.

MPFIL: Multi-Level Protein Feature Integration for Predicting Mutation-Induced Protein-Protein Binding Affinity Changes

MPFIL predicts mutation-induced changes in protein-protein binding affinity by integrating structural and sequential features across multiple levels.

Mimic: A Unified Neural Network Architecture for Modelling Genomic Sequences and Module-level Comparison

This paper introduces Mimic, a unified neural network architecture for genomic sequence modeling, showing it outperforms existing models in protein-DNA interactions and enhancer activity.

Predicting Protein-Protein Interactions Using Vectorized Sequence Descriptors and Machine Learning Classifiers

This paper develops a machine learning framework using sequence descriptors to predict protein-protein interactions, achieving high accuracy with Random Forest and XGBoost models.

AntiOxiPred: Predicting Antioxidant Peptides using Machine Learning Techniques and BLAST

AntiOxiPred predicts antioxidant peptides using machine learning and BLAST, achieving high AUC and MCC on testing and external validation datasets.

flodl-bio: First Rust data, transformers and model layer for entire deep learning on biological sequences

flodl-bio introduces a Rust framework for deep learning on biological sequences, achieving state-of-the-art performance on transcription-factor-binding datasets.

Building from the ground up: augmentation strategies to improve protein language model performance on protein sets

This study investigates augmentation strategies to improve protein language model performance by reducing dataset size and enhancing generalization through active learning and generative models.

Multi-Peptide Prompting Enables In-Context Learning in Protein Language

This paper introduces multi-peptide example prompts (MPEPs) to enable in-context learning in protein language models for peptide classification without training.

Going beyond labels: Semi-supervised learning for enzyme kinetic parameter prediction

This paper improves enzyme kinetic parameter prediction using semi-supervised learning with unlabeled data, showing better performance than labeled data alone.

PHANES: an integrated resource of protein features for the human proteome

PHANES provides a comprehensive, interpretable database of protein features for the human proteome, integrating sequence, structure, and expression data to support biological analysis and machine learning.

In Other Words: Learning When to Trust Generated Protein Function Descriptions

The paper introduces a method to assess the reliability of generated protein function descriptions using a judge model and shows it improves performance across multiple methods.

What does domain-specific pretraining buy? A 170,000-parameter plant peptide language model against the ESM-2 scaling curve

The paper trains a domain-specific plant peptide language model and compares its performance to generalist models, finding domain pretraining provides significant efficiency gains.

Constructed Negatives Shape the Evaluation of Sequence-Based Protein--Protein Interaction Models

The paper evaluates how different constructed negative examples affect sequence-based protein-protein interaction models, showing significant variability in performance metrics based on negative construction methods.

Evidence-resolved benchmarks separate biological context recognition from direct-partner specificity in protein interaction models

The paper benchmarks protein interaction models to distinguish direct partners from non-binding or noncontacting proteins, showing that context recognition and direct specificity are separable capabilities.

DNA-LLM2Vec: Architecture-Aware Task-Conditioned Adaptation of Decoder DNA Foundation Models

DNA-LLM2Vec adapts decoder DNA models to bidirectional context for genomic tasks, improving performance on extended-context tasks but varying by task and backbone.

UPLiFT: Uncertainty Parameterized Activation for Lightweight Fine-Tuning

UPLiFT improves parameter efficiency in fine-tuning protein language models by optimizing activation space instead of network weights, achieving competitive performance with fewer trainable parameters.

Alignment-Aware Genomic Coordinate Encoding for Long-Range DNA Modeling

This paper introduces USM-AA, an alignment-aware genomic foundation model that improves long-range DNA modeling by incorporating reference genome coordinates and reverse-complement masked learning.

Representing biological concepts as directions in foundation model embeddings

This paper shows that biological concepts can be represented as directions in foundation model embeddings, enabling interpretable and reusable analysis of diverse biological properties.

Reformulating Protein Corona Prediction as Universal Protein-conditioned Learning

This paper introduces NanoPCT, a framework that predicts protein corona formation by learning universal patterns across proteins and enables zero-shot prediction for unseen proteins.

CHORUS: A proteome-scale atlas of membrane protein interaction interfaces predicted from sequence

CHORUS predicts membrane protein interaction interfaces from sequence, outperforming existing methods and providing a proteome-scale atlas of interactions.

MINT: Predicting Enzyme-Cofactor Specificity by Multimodal Knowledge Transfer from Protein-Ligand Interactions

MINT predicts enzyme-cofactor specificity using a multimodal framework with transfer learning, achieving high accuracy in cofactor binding prediction.

EdiProPred: Prediction of Edible Plant Tissue-Associated Proteins Using Large Language Models

The paper develops EdiProPred, a model using large language models to classify plant proteins as edible or non-edible based on sequence data, achieving high accuracy.

Mechanistic classification of the AAA superfamily with protein language models

This paper uses a protein language model to classify AAA family proteins by their mechanistic function, showing that sequence context can predict biochemical mechanisms.

AmyloDeep: pLM-based ensemble model for predicting amyloid propensity from the amino acid sequence

AmyloDeep predicts amyloid propensity from amino acid sequences using an ensemble of deep learning models, achieving high accuracy and robustness.

Auditing Protein-Protein Interaction Signals with Sparse Autoencoder Fingerprints

AuditPPI evaluates PPI predictors by transforming language model features into auditable fingerprints, revealing that predictive success does not necessarily reflect biological signal capture.

Quantifying the Provenance-to-Function Gap in Antidiabetic Peptide Prediction: Homology-Aware Evaluation and the ADP-Hybrid Baseline

The paper evaluates the provenance-to-function gap in antidiabetic peptide prediction by re-evaluating benchmarks with homologous peptide grouping and introduces the ADP-Hybrid baseline classifier.

Evidence Scaling for Zero-Shot Protein Reasoning with Large Language Models

This paper shows that adding biological evidence to large language models improves zero-shot protein variant prediction without retraining the model.

GeoEPred: A Multimodal Structure-Aware Geometric Deep Learning Framework for Gram-Negative Bacterial Secreted Effector Prediction with Sequence Semantics

GeoEPred improves effector protein prediction by integrating sequence and structural information to better model functional motifs and spatial conformations.

Repurposing PeTriBERT for Protein Protein Interaction Prediction with Sequence Structure Fusion

The paper presents PeTriPPI2, a hybrid model for protein-protein interaction prediction that combines sequence and structural features, achieving high accuracy on the Pinder dataset.