Machine Learning in Bioinformatics

DualLoc: Full-parameter fine-tuning of cascaded dual transformers for protein subcellular localization prediction

DualLoc improves protein subcellular localization prediction by using full-parameter fine-tuning of dual transformers, achieving better accuracy for multi-compartment localization.

Visualize, Explore, and Select: A protein Language Model-based Approach Enabling Navigation of Protein Sequence Space for Enzyme Discovery and Mining

The paper introduces SelectZyme, a platform using protein language models and dimensionality reduction to visualize and explore protein sequence space for enzyme discovery, enabling candidate selection without relying on sequence identity or annotations.

PepBERT: Lightweight language models for bioactive peptide representation

PepBERT is a lightweight peptide language model that outperforms ESM-2 on most peptide-related tasks, enabling better representation of bioactive peptides for downstream applications.

Glydentify: An explainable deep learning platform for glycosyltransferase donor substrate prediction

Glydentify predicts donor substrates for glycosyltransferases using deep learning, achieving high accuracy and providing explainable insights into enzyme-substrate relationships.

ACENet: a graph neural network for predicting enzyme pHmin using structure and sequence insights

ACENet predicts enzyme pHmin using a graph neural network integrating structural and evolutionary features, achieving high accuracy and stability across conditions.

MoCETSE: A mixture-of-convolutional experts and transformer-based model for predicting Gram-negative bacterial secreted effectors

This paper introduces MoCETSE, a deep learning model that improves prediction of Gram-negative bacterial effector proteins by combining convolutional experts and transformers with relative positional encoding.

Reaction-Conditioned Enzyme Discovery with Multimodal Deep Learning

This paper introduces VenusRXN, a deep learning framework for reaction-conditioned enzyme discovery, achieving high accuracy in identifying enzymes for novel reactions.

A Corporative Language Model for Protein-Protein Interaction, Binding Affinity, and Interface Contact Prediction

This paper introduces a protein-protein language model for predicting interactions, binding affinity, and interface contacts, achieving state-of-the-art results across species and complex systems.

Evolutionary profile enhancement improves protein function annotation for remote homologs

EPERep enhances protein function annotation for remote homologs by using evolutionary input to improve prediction accuracy for underrepresented proteins.

LLPSight: enhancing prediction of LLPS-driving proteins using machine learning and protein Language Models

LLPSight predicts LLPS-driving proteins using machine learning and protein language models, achieving an F1 score of 0.885 and identifying promising targets in the human proteome.

Novel universal domain-centric method for protein classification

The paper introduces a domain-centric method using physico-chemical properties to classify proteins, improving accuracy over traditional phylogeny-based approaches.

Improving Biological Sequence Prediction with AlphaFold2 Representation

The paper evaluates AlphaFold2's representation for improving sequence-based prediction tasks, showing superior performance over traditional features.

Influ-BERT: A Domain-Adaptive Genomic Language Model for Advancing Influenza A Virus Research

Influ-BERT improves IAV subtype identification by using domain-adaptive pretraining and achieves high F1-scores across underrepresented subtypes.

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

This paper introduces a deep learning model that predicts if two proteins catalyze the same reaction using ESM-C latent representations and attention mechanisms, outperforming existing methods.

Leveraging learned representations and multitask learning for lysine methylation site discovery

This work introduces a transformer-based model for lysine methylation site prediction, achieving state-of-the-art accuracy through multitask learning and validation with mass spectrometry.

A General Framework for Injecting BiophysicalPriors into Protein Embeddings

This paper introduces a framework that integrates biophysical priors into protein embeddings to improve DeltaDeltaG prediction, showing that general-purpose encoders can outperform specialized models.

CrossAffinity: A Sequence-Based Protein-Protein Binding Affinity Prediction Tool Using Cross-Attention Mechanism

CrossAffinity predicts protein-protein binding affinity using a sequence-based cross-attention model, outperforming both structure-based and sequence-based methods on multiple test sets.

Guided tokenization and domain knowledge enhance genomic language models' performance

This paper introduces guided tokenization and domain adaptation to improve genomic language models' performance in classification and annotation tasks.

Enzyme Classification via Semi-Supervised Functional ResidueLearning

This paper introduces SLEEC, a semi-supervised framework for enzyme classification that achieves state-of-the-art performance and provides residue-level annotations.

A flaw in using pre-trained pLLMs in protein-protein interaction inference models

This paper identifies data leakage in pre-trained pLLMs used for protein-protein interaction inference and shows their limitations in generalization.

A Systematic Benchmark for Peptide Property Prediction

This paper introduces PPB, a benchmark for peptide property prediction, showing that large pre-trained models outperform others and highlighting challenges in clustering short peptides.

Prediction of bacterial protein-compound interactions with only positive samples

This paper introduces BIN-PU, a PU learning framework for predicting bacterial protein-compound interactions using only positive samples, showing improved performance over existing methods.

SIP-BERT: A multi-organism deep strategy for predicting self interaction in proteins

SIP-BERT predicts self-interacting proteins using organism-specific datasets, outperforming existing methods and generalizing to under-annotated species.

Automatically Defining Protein Words for Diverse Functional Predictions Based on Attention Analysis of a Protein Language Model

The paper introduces a toolkit using protein language models to define 'protein words' for functional prediction, outperforming motif-based methods across multiple tasks.

EvoPool: Evolution-Guided Pooling of Protein Language Model Embeddings

EvoPool improves variant effect prediction by integrating evolutionary information into protein language model embeddings using optimal transport.

Predicting evolutionary rate as a pretraining task improves genome language model representations

This paper introduces a pretraining task for genome language models that predicts evolutionary rate, showing improved performance over sequence-only pretraining.

ProtLoc-GRPO: Cell line-specific subcellular localization prediction using a graph-based model and reinforcement learning

This study introduces ProtLoc-GRPO, a reinforcement learning method that improves subcellular localization prediction by optimizing PPI networks, achieving a 7% macro-F1 score improvement.

An AI-driven pipeline for the discovery of hidden peptides in plant proteomes: the CLE family as a case study

The paper introduces an AI-driven pipeline using Protein Language Models to discover hidden CLE peptides in plant proteomes, achieving high classification accuracy and identifying previously unannotated candidates.

Decoding protein-membrane binding interfaces from surface-fingerprint-based geometric deep learning and molecular dynamics simulations

This paper introduces MaSIF-PMP, a geometric deep learning model for predicting protein-membrane binding sites, showing improved performance and validating predictions with molecular dynamics simulations.

GNN2Pfam: Integrating protein sequence and structure with graph neural networks for Pfam domain annotation

GNN2Pfam uses graph neural networks to improve Pfam domain annotation by integrating protein sequence and structure data, outperforming HMM-based methods.