Nishanth Basava | Searcharxiv

arXiv2025

bayesNMF: Fast Bayesian Poisson NMF with Automatically Learned Rank Applied to Mutational Signatures

Bayesian Poisson Non-Negative Matrix Factorization (NMF) is widely used to model count data, including in cancer mutational signature analysis. However, standard Gibbs samplers rely on computationally expensive Poisson augmentation, and current software implementations learn the latent rank either through slow and potentially subjective heuristic rank selection or with automatic approaches that do not report posterior uncertainty. In this paper, we introduce bayesNMF, an MH-within-Gibbs sampler to address both of these limitations. First, we define high-overlap proposals for Metropolis-Hastings sampling to remove the need for Poisson augmentation. Second, we define a BIC-based sparsity prior to learn rank automatically within the Bayesian formulation while allowing for posterior uncertainty quantification. We provide an open-source R software package with all of the models and plotting capabilities demonstrated in this paper on GitHub at jennalandy/bayesNMF. Although our applications focus on cancer mutational signatures, our software and results can be extended to any use of Bayesian Poisson NMF.

stat.ME↗

arXiv2024

Revolutionizing Personalized Cancer Vaccines with NEO: Novel Epitope Optimization Using an Aggregated Feed Forward and Recurrent Neural Network with LSTM Architecture

As cancer cases continue to rise, with a 2023 study from Zhejiang and Harvard predicting a 31 percent increase in cases and a 21 percent increase in deaths by 2030, the need to find more effective treatments for cancer is greater than ever before. Traditional approaches to treating cancer, such as chemotherapy, often kill healthy cells because of their lack of targetability. In contrast, personalized cancer vaccines can utilize neoepitopes - distinctive peptides on cancer cells that are often missed by the body's immune system - that have strong binding affinities to a patient's MHC to provide a more targeted treatment approach. The selection of optimal neoepitopes that elicit an immune response is a time-consuming and costly process due to the required inputs of modern predictive methods. This project aims to facilitate faster, cheaper, and more accurate neoepitope binding predictions using Feed Forward Neural Networks (FFNN) and Recurrent Neural Networks (RNN). To address this, NEO was created. NEO requires next-generation sequencing data and uses a stacking ensemble method by calculating scores from state-of-the-art models (MHCFlurry 1.6, NetMHCstabpan 1.0, and IEDB). The model's architecture includes an FFNN and an RNN with LSTM layers capable of analyzing both sequential and non-sequential data. The results from both models are aggregated to produce predictions. Using this model, personalized cancer vaccines can be produced with improved results (AUC = 0.9166, recall = 91.67 percent).

cs.LG↗