Designing and building large scale machine learning systems for medical coding.

Fathom Health is a deep learning NLP system to accelerate medical reimbursement, backed by world class investors including Founders Fund, Google Ventures, 8VC, and Stanford, as well as founders and early employees from companies like Google, Dropbox, Airbnb, and athenahealth.

• Designed Machine Learning workflow for Convolutional Neural Network(CNN) based compound fingerprint prediction for metabolite annotation project (fully conducted in Python), improved top 1 ranking accuracy by 35.7% compared with other tools. (Manuscript in preparing.)

• Performed data parsing, preprocessing for mass spectrum data from NIST20 and MONA databases which including 600,000+ mass spectrum from 26000+ compounds; built ML models for fingerprint prediction; evaluated models by multi approach in Python.

• Constructed data visualization and correlation analysis between mRNA and miRNA and between mRNA and metabolites for Hepatocellular Carcinoma in R. Was the co-author of the paper published in IEEE EMBC 2020.

• Led a team consists of 4 Jr. Research Assistants and collaborated with other members in Resson Lab and reported project progress to manager at weekly meetings.

Research Practicum based on Statistical Machine Learning in Longitudinal Studies

• Extracted the longitudinal data including 9 in-person examinations during 30 years from the Coronary Artery Risk Development in Young Adults (CARDIA) study, which enrolled 5115 healthy Black and White men and women who were aged 18 to 30 years from 1985 to 1986 in 4 US urban areas.

• Applied mixed statistical machine learning methods to show the potential risk factors for cardiovascular disease events.

• Attended weekly project meetings with teams from different organizations to discuss progress and plan for multi research projects.

• Completed data cleaning for company name disambiguation from 3 different databases with 17300+ items using vlookup and Google Sheets, providing directly usable data to data team.

• Created selection methods using PermID as new primary key to union 3 databases without duplication.

• Conducted information collection on 100+ types of drugs such as monoclonal antibody, fusion protein, pd-1, etc., to provide data support for drug research and development concerned by Henlius Biotech.

• Implemented weekly product data specification standardization of about 12000+ items, and regularly updated molecular target information for new biological drugs in the database.

• Assisted in collecting more than 160 financial statements of global pharmaceutical companies from 2017 to 2018, and provided comparative analysis data.