Ultra-Fast Combinatorial Analysis for Biomarker Discovery

Project Summary Abstract SignificanceOne of healthcare s largest challenges is that many patients do not respond to treatmentBy recent estimatesaroundof drugs are effective for less thanof patientsThis causes enormous physicalsocialand economic sufferingThe annual cost of ineffective treatment is estimated at $billion per year in the US aloneMoreovervariable treatment response contributes to the rising cost of drug developmentcurrently around $billion per drugThe fundamental reasons that treatment responses vary is that there are many different cell types in peopleand cell types vary person to personIn the computational world this problem of complexity is calleddimensionality curseTo avoid the enormous computational burden of high dimensionalityresearchers developed deep learning and dimensionality reduction tools like T SNEPCAHowever in the process of solving computational challengescrucial biological interpretability is lost leading to a bigger issuereliability of these methodsHypothesisWe hypothesized that faster combinatorics algorithms combined with massive parallel computing techniques would make computation solutions for single cell high dimensional data tractablePreliminary DataWe have demonstrated our approach algorithm on a publicly available AIDS datasetOur method produced biologically interpretableaccurate results at an ultra fast speedx faster vscurrent methodsmaking the approach tractableThe approach was further validated for biological significance and accuracy through collaboration with an NYU Medical Center researcherwho has now joined the team as Chief Scientific AdvisorSpecific AimsThis project entails AimBuilding a production gradecomputationally efficient and scalable system for creation and storage of all combinatorial phenotypic signaturesincluding frequency of each cell typeby patientwith associated metadataAimIntegrating an automated system for comparing phenotypic frequencies across patient groupsincorporating standard biostatistical analyses and machine learning based modelingAimApplying our new tool to multiple existinghigh parameter cancer and infectious disease datasets to demonstrate the value of combinatorics over dimension reduction for accurately predicting clinical outcomes at ultra fast speedTogetherthese studies will demonstrate a new computational approach to high dimensional single cell data without losing biological information or interpretability at an ultra fast speedrevolutionizing precision medicine Project Narrative Modern technology to measure cells has improved significantlyenabling clinicians and researchers to precisely identify cell typesbiomarkerthat predict successful treatmentHoweverdata analysis tools have lagged behindmaking it harder to use modern technology for precision medicineA novel computational approach to analyze large volumes of dataquickly and in a fully interpretable way to predict treatment outcomesis a critical need that can unleash the benefits of modern instruments and technologies