Murray Patterson

The paper introduces a compression and k-mer based method for anti-cancer peptide classification, achieving better performance than deep learning models without requiring custom features or pre-trained models.

This paper improves epitope prediction using graph convolutional networks and protein language model embeddings, achieving an AUCROC of 0.65 on a benchmark dataset.

This paper introduces DeepPWM-BindingNet, a deep-learning framework that improves DNA-protein binding prediction by integrating PWM features with CRNNs and attention mechanisms.

This paper introduces Poincare distance measures for genomic and proteomic sequence analysis, showing improved performance in classification tasks compared to existing methods.

This study develops a model combining protein language models and machine learning classifiers to predict protein-DNA binding sites, showing improved performance on benchmark datasets.

This paper benchmarks ML models for classifying COVID-19 spike sequences, finding that deep neural networks and k-mer features outperform traditional methods in accuracy and robustness.

This paper evaluates different kernels for t-SNE to improve its performance on biological sequencing data, finding that the cosine similarity kernel provides the best results for SARS-CoV-2 spike protein sequences.

This paper introduces an efficient sequence embedding method for SARS-CoV-2 variant classification, improving predictive performance and computational efficiency over existing kernel methods.

Hist2Vec is a kernel-based embedding method for biological sequence classification that outperforms existing methods with >76% and >83% accuracy on DNA and protein datasets, respectively.

The paper develops a pipeline called plastic for tumor phylogeny analysis, integrating data simplification, phylogeny inference, and comparison with improved speed and visualization.

This paper clusters SARS-CoV-2 sequences to identify variants and subtypes, showing improved performance over existing methods and detecting rare variants in large datasets.

The paper introduces SASC, a method for inferring cancer progression from single-cell sequencing data that allows mutation losses, showing high accuracy on simulated and real datasets.

HapCHAT improves haplotype assembly by adapting error thresholds for higher coverage long reads, enhancing accuracy and reducing computational resources.

The paper analyzes metabolic function evolution across 1459 species to identify co-evolving metabolic modules using phylogenetic and network analysis.

Re-aligning reads to both alleles gives WhatsHap the lowest PacBio switch error rates (full coverage: 0.24% real, 0.02% simulated) and better indel phasing than hapCUT and phASER.