Alexander Spyridon Baras

This paper quantifies how much V and J gene identity can be recovered from TCR CDR3 sequences using generative models and shows that J genes are nearly determined while V genes are partially recoverable with group-level structure.

TESSERA is a foundation model for the cancer genome that uses self-supervised learning to predict variant pathogenicity, tumour typing, and treatment response, revealing interpretable biomarkers for chemotherapy selection.

This paper applies multi-instance learning to automate flow cytometry analysis, demonstrating effectiveness in cancer subtyping, HIV survival, AML minimal residual disease, and COVID-19 severity assessment.

This study shows that a pretrained tissue patch encoder can effectively encode cellular graphs, capturing cell-level features and spatial relationships in tissue patches.

This study introduces a two-cutoff approach and neural networks to model non-monotonic relationships between tumor mutational burden and clinical outcomes, finding that non-monotonic relationships can exist.

This study evaluates a recombinant BCG strain that releases a STING agonist, showing improved antitumor efficacy in bladder cancer models compared to wild-type BCG.

The paper introduces a deep learning framework for genomic data that aggregates and predicts somatic mutations and other sparse genomic measures using attention-based multiple instance learning.

This paper introduces probabilistic mixture models to calibrate panel-derived tumor mutational burden (TMB) to exomic TMB, showing improved accuracy over linear regression.

The paper estimates the true mutation VAF distributions from low-depth sequencing by comparing with high-depth data, finding TCGA MC3 underreports mutations by 17%.

DeepTCR uses deep learning to analyze T-cell receptor sequences, improving classification and prediction in immunology and clinical settings.

The paper applies transfer learning using ResNet50 to extract diagnostic features from digital pathology images of breast lesions at various magnifications for classification into histological labels.

AI-MHC improves Class I and Class II HLA-binding predictions using a deep learning framework that integrates allele information and achieves state-of-the-art performance.

miRge 2.0 improves miRNA analysis with enhanced detection, A-to-I editing analysis, and better alignment, outperforming existing tools on human datasets.

The paper establishes the landscape of human cell-specific microRNA expression using small RNA-seq data from diverse cell types and tissues, identifying novel microRNA loci and patterns correlated with super-enhancer activity.