Machine Learning in Bioinformatics

Prediction of protein-carbohydrate binding sites from protein primary sequence

This paper introduces StackCBEmbed, a machine learning model that predicts protein-carbohydrate binding interactions from primary sequence, achieving better performance than existing methods.

DeepLocRNA: An Interpretable Deep Learning Model for Predicting RNA Subcellular Localization with domain-specific transfer-learning

DeepLocRNA predicts RNA subcellular localization using transfer learning from pre-trained RBP-binding models, achieving state-of-the-art performance across human and mouse RNA.

TopEC: Improved classification of enzyme function by a localized 3D protein descriptor and 3D Graph Neural Networks

TopEC improves enzyme function classification using 3D graph neural networks and a localized 3D descriptor, achieving high accuracy in EC class prediction.

DeepGOMeta: Predicting functions for microbes

DeepGOMeta predicts protein functions for microbes using a deep learning model trained on microbial data, improving function prediction for novel proteins.

ProtHyena: A fast and efficient foundation protein language model at single amino acid resolution

ProtHyena is a foundation protein language model that efficiently processes long protein sequences at the single amino acid level with reduced computational complexity.

OmniNA: A foundation model for nucleotide sequences

OmniNA is a foundation model for nucleotide sequences trained on diverse species data, achieving performance across 17 tasks and offering insights into mutation effects on DNA/RNA processing.

TransMEP: Transfer learning on large protein language models to predict mutation effects of proteins from a small known dataset

TransMEP uses transfer learning with Gaussian process regression to predict mutation effects from small datasets, showing improved performance over previous methods in protein engineering.

EquiPNAS: improved protein-nucleic acid binding site prediction using protein-language-model-informed equivariant deep graph neural networks

EquiPNAS improves protein-nucleic acid binding site prediction using a pLM-informed equivariant graph neural network, outperforming existing methods on multiple datasets.

Detecting Anomalous Proteins Using Deep Representations

This paper uses deep learning to detect anomalous proteins by identifying unusual properties in function, phylogenetic families, and sequence segmentation.

AAclust: k-optimized clustering for selecting redundancy-reduced sets of amino acid scales

AAclust optimizes k-values for clustering amino acid scales to reduce redundancy, improving performance on protein prediction benchmarks.

TemBERTure: Advancing protein thermostability prediction with Deep Learning and attention mechanisms

This paper introduces TemBERTure, a Deep Learning framework for predicting protein thermostability using sequence data, showing improved performance with diverse datasets and attention mechanisms.

Seq-InSite: sequence supersedes structure for protein interaction site prediction

Seq-InSite predicts protein interaction sites using sequence data, outperforming sequence-based models and matching structure-based predictors.

ProSST: Protein Language Modeling with Quantized Structure and Disentangled Attention

ProSST integrates protein sequences and structures into a Transformer-based model, achieving state-of-the-art performance in mutation effect prediction and downstream tasks.

Multi-Modal CLIP-Informed Protein Editing

This paper introduces ProtET, a method for CLIP-informed protein editing using multi-modality learning, achieving significant improvements in protein stability and functionality.

M-Ionic: Prediction of metal ion binding sites from sequence using residue embeddings

M-Ionic predicts metal ion binding sites from protein sequences using residue embeddings, achieving high accuracy for known ions and extending predictions to additional ions.

Improving protein function prediction by learning and integrating representations of protein sequences and function labels

The paper introduces TransFew, a transformer model that integrates protein sequence and function label representations to improve protein function prediction, especially for rare function terms.

RUDEUS, a machine learning classification system to study DNA-Binding proteins

RUDEUS is a machine learning system that identifies DNA-binding proteins and classifies their DNA interaction type, achieving high precision in both tasks.

Learning Peptide Properties with Positive Examples Only

The paper develops a semi-supervised deep learning method using only positive examples to predict peptide properties like solubility and antimicrobial activity.

Multimodal Pretraining for Unsupervised Protein Representation Learning

This paper introduces a symmetry-preserving multimodal pretraining framework for proteins, achieving competitive results in multiple tasks using a PyTorch implementation.

DLKcat cannot predict meaningful kcat values for mutants and unfamiliar enzymes

This paper evaluates the DLKcat model's ability to predict kcat values for enzymes and mutants, finding it unreliable for low sequence identity and poor at capturing mutation effects.

ESM-PVP: Identification and classification of phage virion proteins with a large pretrained protein language model and an MLP neural network

ESM-PVP uses a pretrained protein language model and an MLP to identify and classify phage virion proteins, outperforming previous methods on a new benchmark dataset.

Embedding-based alignment: combining protein language models and alignment approaches to detect structural similarities in the twilight-zone

This paper introduces an embedding-based alignment method using protein language models to detect structural similarities in the twilight zone, outperforming existing approaches without training.

GAPS: Geometric Attention-based Networks for Peptide Binding Sites Identification by the Transfer Learning Approach

GAPS uses geometric attention mechanisms and transfer learning to identify protein-peptide binding sites, achieving state-of-the-art performance.

Pre-training with pseudo-labeling compares favorably with large language models for regulatory sequence prediction

The paper introduces a semi-supervised learning approach for regulatory sequence prediction using pseudo-labeling, showing strong performance improvements over supervised learning and comparable results to DNABERT2.

Systematic Investigation of Machine Learning on Limited Data: A Study on Predicting Protein-Protein Binding Strength

This study develops and evaluates machine learning models for predicting protein-protein binding affinity using limited data, achieving significant improvements over existing predictors.

A joint embedding of protein sequence and structure enables robust variant effect predictions

This paper introduces SSEmb, a model that integrates protein sequence and structure to improve variant effect predictions and downstream tasks like binding site prediction.

Structure-Infused Protein Language Models

This paper shows that incorporating protein structure information into protein language models improves their performance on structure-related tasks without requiring structure access during inference.

Enhancing Recognition and Interpretation of Functional Phenotypic Sequences through Fine-Tuning Pre-Trained Genomic Models

This paper fine-tunes pre-trained genomic models on genotype-phenotype datasets to improve DNA sequence classification, achieving high accuracy on the HERV dataset.

DeepNeuropePred: a robust and universal tool to predict cleavage sites from neuropeptide precursors by protein language model

DeepNeuropePred predicts neuropeptide cleavage sites using a protein language model and CNNs, outperforming existing methods across species.

Prediction of Interactions between Cell Surface Proteins by Machine Learning

This paper develops a machine learning framework to predict interactions between cell surface proteins, achieving over 70% accuracy in human proteins and validating results against experimental data.