Juan Manuel Berros

I analyzed the distribution of several genetic disease scores for my Ph.D. This project has tested my knowledge of statistics acquired in college and my coding capabilities from years of expertise in the private sector. It also required heavy use of Python's data science stack to achieve the showcased results.

While at Biocodices (medical genetics laboratory), I built a complex pipeline of genetic data processing based on the Python’s workflow framework Luigi (a framework similar to Airflow, developed at Spotify).

The workflow started with the raw output of various DNA Sequencers (i.e. machines that read the sequence of DNA samples) and went through numerous standard bioinformatic quality assurance and filtering steps until a subset of medically relevant genetic mutations was kept for reporting.

For this project, the client had a data challenge to quickly analyze a dataset of sites where cargo accumulates over time and is picked up according to different signals—a mixture of automatic and scheduled pickups.

I developed from scratch an internal web application based on Ruby on Rails for a medical genomics laboratory and maintained it for over five years.

The app serves as an online inventory of patients, clinics, doctors, and DNA testing results and implements the CRUD operations through a user-friendly interface. It also automates the data entry of forms and allows bulk editions of different types of data. It generates daily stats of the lab's results that can be interactively filtered and customized. It also automates the generation and interactive edition of PDF reports to communicate the genetic testing results to clinics.

The maintenance of this app required strict adherence to good coding style and software architecture practices and the development of an exhaustive testing suite.

I worked at a laboratory that was part of the Proyecto-PAIS, the country-wide effort in Argentina to sequence and analyze SARS-Cov-2 DNA samples. I explored the performance of the DNA sequencing and the mutations found in the Coronavirus genome samples.

Cleaned up a database of millions of company names by addressing multiple parsing issues. To decide whether a company name needed revision, I developed a quantitative score that captured the intuition of what a "correct" company name was in that context, leveraging on available company data like social media handles, the domain, and available names from different sources.

I performed extensive data analysis and implemented an OOP solution to score geolocation API results and decide whether the result was trustworthy or not. The evaluation was productionized and performs thousands of daily evaluations.