About the Laboratory

We create the cutting-edge deep learning technologies for mass spectrometry data analysis to be used in life sciences and biomedical applications. We are an international, English-speaking, multidisciplinary group with background in computer science, math, and molecular biology.

Tandem Mass Spectrometry is the de facto method to identify molecules in complex mixtures (e.g. blood, cell, food) in many areas like:

• proteomics to identify proteins in biological samples such as blood;
• clinical applications to identify proteins related to cancer or other diseases;
• pharmaceutical analysis to determine the molecular effects of new drugs;
• environmental contamination analysis to ensure that the air, drinking water, soils, and food are safe to consume and does not contain pollution, heavy metals, hormone, pesticides, and herbicides; forensic analysis to trace of evidence in arson investigation, drug abuse;
• metabolomics to identify small molecules used by bacteria for communication in microbiome.

Project #1. A computational annotation of mass spectrum data can be either correct or incorrect, but it cannot be verified by human curators. We develop and investigate randomized tests to assign accurate confidence scores to annotations.

Project #2. Score functions are the work-horses in the spectrum annotation methods. We develop deep learning methods for score functions with better discriminative power via utilizing explainable information learned from spectrum data.

Project #3. Complex deep learning models having millions of parameters can discover subtle clues in the data to provide seemingly correct prediction, albeit without learning a proper generalization or without a human-like reasoning process. This pitfall is known as “clever Hans” phenomenon. We develop and investigate neuro-symbolic methods, differentiable reasoning technologies with end-to-end learning fashion to learn to reason in general data and to reason and explain spectrum data annotations.