Machine Learning in Bioinformatics

PARROT is a deep learning framework for analyzing large protein datasets, capable of classification and regression tasks using only raw protein sequences.

IAV-CNN predicts influenza A virus antigenic variants using a 2D convolutional neural network with a new amino acid representation, achieving state-of-the-art performance on three datasets.

TripletProt improves protein functional annotation using Siamese networks, achieving better F1 scores than existing methods in sub-cellular localization and gene ontology prediction.

This paper introduces ABLE, an attention-based bidirectional-LSTM model for enzyme classification using protein FASTA sequences, achieving an F1-score of 0.834.

The paper introduces improved datasets and evaluation methods for DNA-binding protein prediction, showing that a gradient boosting model and a BLAST-based baseline outperform previous methods.

DeepLncLoc predicts lncRNA subcellular localization using a subsequence embedding method and text CNN, outperforming k-mer-based models and existing predictors.

The paper introduces DeepVHPPI, a deep learning framework for predicting virus-host protein interactions for novel virus sequences, showing significant improvements over existing methods for SARS-CoV-2, H1N1, and Ebola.

The paper develops deepredeff, a deep learning model ensemble for classifying plant pathogen effector proteins from sequence data, achieving higher accuracy and lower false positives than existing tools.

The paper predicts human-virus protein-protein interactions using transfer learning with CNN and MLP, outperforming existing methods and applying it to SARS-CoV-2.

The paper introduces MASSP, a deep-learning system that simultaneously predicts secondary structure, transmembrane segments, and topology of proteins, outperforming existing methods in accuracy and applicability.

This paper trains and evaluates language models on protein sequences to predict structural and localization features, achieving state-of-the-art results without evolutionary information.

In-Pero predicts peroxisomal protein sub-localisation using deep learning embeddings, achieving over 0.9 classification accuracy.

This paper introduces evotuning protocols for Transformer-based variant effect prediction, showing improved performance by incorporating domain architecture information from homologous sequences.

The paper improves protein structure classification by combining GmdRQA and HVG with reduced computational cost compared to state-of-the-art methods.

This paper introduces a Graph Convolutional Network for predicting protein function and identifying important residues, showing improved accuracy over sequence-based methods.

The paper introduces LSTM-PHV, a sequence-based model predicting human-virus protein-protein interactions using LSTM and word2vec, achieving high accuracy on imbalanced data.

The paper introduces a method combining CNNs and GCNs to predict paratope residues in antibodies by integrating sequence and spatial information.

The paper develops a CNN-based tool to identify 6mA sites in rice genomes and shows it generalizes to other plant genomes.

The paper improves the purity of protein functional families by clustering them using sequence embeddings, doubling the number of pure clusters per family.

ABC-finder identifies and predicts the topology of ABC proteins across organisms using a Docker-based web server.

SubFeat predicts functionalities of DNA, RNA, and protein sequences using an ensemble classifier, outperforming existing methods on four datasets.

This paper presents a model for predicting fertility-related proteins using multi-information fusion and LightGBM, achieving high accuracy in cross-validation and independent testing.

DeepDigest predicts proteolytic cleavage sites for eight proteases using a deep learning model, outperforming traditional methods across multiple datasets.

The paper trains a deep language model on 250 million protein sequences to learn biological representations that capture structural and functional properties.

TALE is a deep learning model for protein function annotation that uses joint sequence-label embeddings to improve generalizability to novel sequences, species, and functions.

This paper introduces a deep forest-based method for predicting protein-protein interactions, achieving high accuracy in yeast and Helicobacter pylori and improving drug discovery.

AIRBP uses machine learning to accurately identify RNA-binding proteins from sequence data, outperforming existing methods across multiple species.

The paper develops a structure-aware model using graph convolutional networks and predicted contact maps to predict protein solubility from sequence, showing improved performance over existing methods.

DNABERT is a pre-trained transformer model for DNA sequences that achieves state-of-the-art performance on sequence prediction tasks and enables visualization of nucleotide-level importance across species.

The paper introduces dom2vec, a method for learning protein domain embeddings and evaluates their quality using intrinsic strategies, showing they perform well on downstream tasks.