Machine Learning in Bioinformatics

Diverse Genomic Embedding Benchmark for functional evaluation across the tree of life

The paper introduces DGEB, a benchmark for evaluating genomic embeddings across diverse sequences and functions, revealing performance limitations in existing models.

TUnA: An uncertainty aware transformer model for sequence-based protein-protein interaction prediction

TUnA is a Transformer model that predicts protein-protein interactions from sequence data and provides uncertainty estimates for unseen sequences, improving reliability.

Predicting Enzyme Functions Using Contrastive Learning with Hierarchical Enzyme Structure Information

This paper introduces EnzHier, a machine learning algorithm that improves enzyme function prediction by integrating sequence, structural information, and functional hierarchy, achieving better accuracy than existing tools.

PlasGO: enhancing GO-based function prediction for plasmid-encoded proteins based on genetic structure

PlasGO predicts GO terms for plasmid-encoded proteins using a hierarchical model, achieving high precision in function annotation.

Enhancing predictions of protein stability changes induced by single mutations using MSA-based Language Models

The paper improves prediction of protein stability changes from single mutations using MSA-based language models and a curated dataset, achieving better performance than existing methods.

Predicting protein functions using positive-unlabeled ranking with ontology-based priors

This paper introduces PU-GO, a method for protein function prediction that addresses the positive-unlabeled ranking problem using ontology-based priors, showing improved performance on benchmark datasets.

AntiCPs-CompML: A Comprehensive Fast Track ML method to predict Anti-Corona Peptides

AntiCPs-CompML predicts anti-coronavirus peptides using machine learning, achieving high accuracy and efficiency in identifying potential therapeutic candidates.

SCREEN: a graph-based contrastive learning tool to infer catalytic residues and assess mutation tolerance in enzymes

SCREEN is a graph neural network that predicts catalytic residues and assesses mutation tolerance in enzymes by integrating structural and functional information.

MetaLATTE: Metal Binding Prediction via Multi-Task Learning on Protein Language Model Latents

MetaLATTE predicts metal binding in proteins using multi-task learning on ESM-2 embeddings, enabling classification of metal-binding properties without requiring structural input.

CD-GPT As a Biological Foundation Model Bridging the Gap between Molecular Sequences Through Central Dogma

CD-GPT is a foundation model that captures sequence relationships between DNA, RNA, and proteins, performing well in predictive and generative tasks across molecular types.

Protein-protein interaction prediction via structure-based deep learning

This study introduces RSPPI, a deep learning framework for predicting protein-protein interactions using structural and sequence data, showing improved cross-species performance over existing methods.

VIPER: A General Model for Prediction of Enzyme Substrates

VIPER improves enzyme substrate prediction by 30% over existing models and introduces a high-quality dataset to address previous limitations.

Prediction of protein functions using Semantic Based Regularization

The paper predicts protein functions using a semantic-based regularization kernel method with logic rules, showing improved performance in bioinformatics applications.

Democratizing Protein Language Models with Parameter-Efficient Fine-Tuning

This paper applies parameter-efficient fine-tuning methods to proteomics, achieving competitive or better performance than traditional fine-tuning with significantly reduced computational resources.

Training Compute-Optimal Protein Language Models

This paper investigates optimal training of protein language models by analyzing the relationship between model size, training data, and performance, finding that diverse training data and scaling laws improve results.

Identification of Family-Specific Features in Cas9 and Cas12 Proteins: A Machine Learning Approach Using Complete Protein Feature Spectrum

The paper uses machine learning to identify distinguishing features of Cas9 and Cas12 proteins, finding specific tripeptides important for their function.

Protein sequence domain annotation using a language model

PSALM annotates protein domains using a language model, achieving sensitivity-specificity tradeoff comparable to HMMER on a large benchmark.

Accelerating High-Performance Classification of Bacterial Proteins Secreted via Non-Classical Pathways: no needing for deepness

This paper develops a fast machine learning method to classify non-classical secreted bacterial proteins with 91% accuracy, outperforming existing methods in speed without sacrificing accuracy.

FusOn-pLM: A Fusion Oncoprotein-Specific Language Model via Focused Probabilistic Masking

This paper introduces FusOn-pLM, a protein language model fine-tuned on fusion oncoprotein sequences using a novel masked language modeling strategy, improving performance on specific benchmarks and disorder prediction tasks.

Deciphering Peptide-Protein Interactions via Composition-Based Prediction: A Case Study with Survivin/BIRC5

The paper predicts peptide-protein interactions using a composition-based model, showing that electrostatic and primary structure features are insufficient for accurate predictions.

Prot2Token: A multi-task framework for protein language processing using autoregressive language modeling

Prot2Token is a multi-task framework that uses autoregressive language modeling for protein prediction tasks, achieving performance comparable to specialized models and enabling integration with large language models.

ProtEx: A Retrieval-Augmented Approach for Protein Function Prediction

ProtEx improves protein function prediction by using retrieved exemplars, achieving state-of-the-art results for EC numbers, GO terms, and Pfam families.

Kermut: Composite kernel regression for protein variant effects

Kermut is a Gaussian process regression model for protein variant effect prediction that provides uncertainty estimates through its posterior.

Prediction of peptide hormones using an ensemble of machine learning and similarity-based methods

The paper develops a hybrid model combining machine learning and similarity-based methods to predict hormonal peptides, achieving high accuracy on a dataset from plants and animals.

iResNetDM: Interpretable deep learning approach for four types of DNA modification prediction

This paper introduces iResNetDM, a deep learning model that predicts four types of DNA modifications with high performance and interpretability.

ThermoFinder: A sequence-based thermophilic proteins prediction framework

ThermoFinder is a sequence-based framework for predicting thermophilic proteins that outperforms existing tools on benchmark datasets and shows strong performance on newly constructed datasets.

Cramming Protein Language Model Training in 24 GPU Hours

This paper trains a protein language model in 24 GPU hours, achieving performance comparable to a much larger model trained for 15,000x more GPU hours.

CLEAN-Contact: Contrastive Learning-enabled Enzyme Functional Annotation Prediction with Structural Inference

The paper introduces CLEAN-Contact, a contrastive learning framework for enzyme function prediction using amino acid sequences and contact maps, showing improved accuracy over existing methods.

ProteinCLIP: enhancing protein language models with natural language

ProteinCLIP refines pre-trained protein language model embeddings using contrastive learning between amino acid sequences and functional text, achieving state-of-the-art results in protein tasks.

Predicting transcriptional activation domain function using Graph Neural Networks

The paper predicts transcriptional activation domain function using graph neural networks, achieving high accuracy and outperforming existing methods.