Susanna Morin | Data Scientist

Hi, I'm Susanna Morin.

A self-driven, quick starter, passionate data scientist with 2 years of experience turning complex data into clear insights and actionable strategies with the goal of patient care optimization.

About

I hold a strong foundation in predictive and descriptive analytics, as well as statistical modeling applied to Electronic Health Records and Claims Data.

I've successfully led projects for government-funded population healthcare programs that have informed policy and resulted in additional funding at a state and federal level.

My main tech stack relies heavily on using Python, R, and SQL to ingest, process, analyze, and visualize my data.

Languages: Python, R, C++, HTML/CSS, Bash
Databases: MySQL, PostgreSQL
Libraries: NumPy, Pandas, Matplotlib, ggplot2, lm
Tools & Technologies: Git, JIRA, TortoiseSVN

I’m particularly drawn to opportunities that allow me to work collaboratively with key business stakeholders to identify areas of value, develop solutions, and deliver insights to reduce overall cost of care for members and improve their clinical outcomes.

Experience

Gainwell Technologies

Data Scientist

Built a survival analysis model using the cox proportional hazards regression to investigate the association between the survival time in the Severe and Persistently Mentally Ill (SPMI) patients and predictor variables of interest with the potential to inform policy in healthcare
Maintained and created tables used in downstream population healthcare analysis using Electronic Health Records (EHRs)
Conducted a total cost of care time series analysis on healthcare provider utilization trends
Developed and validated a Generalized Linear Mixed Effects Model for utilization care costs
Tools: Python, R, SQL, Azure, Tableu, TortoiseSVN

November 2022 - Present | Remote, USA

UCSF Health

Clinical Informatics Analyst

Conducted data validation in collaboration with Nima Aghaeepour Laboratory at Stanford University; Focused research efforts on patient phenotyping and trajectory prediction in neonatal health and morbidity resulting in a publication in Science Translational Medicine
Pre-processed data using R-Studio; Analyzed relationships between women’s health factors and offspring health outcomes in pre-term labor using statistical methods
Generalized a CNN model measuring knee osteoarthritis and improved performance by changing the biomarker from bone shape to cartilage thickness

Tools:

July 2021 - December 2022 | San Francisco, CA

UCSC Genomics Institute

Algorithm Developer

Conducted research contributing to the development and driving of technical standards for genomic data; Focused on infrastructure for graph-based genomics
Developed a genotyper using the Markov Chain Monte Carlo probabilistic model that supports standard variant calling formats; Improved accuracy and performance of genotyper using the Min-Cut algorithm to break out of sampling bottlenecks maximizing mixing efficiency
Established evaluation methods that compare accuracy metrics against gold-standard datasets
Tools: C++, Git, Bash, Statistical Inference

July 2019 - May 2021 | Santa Cruz, CA

Gladstone Institutes

Data Scientist Intern

Developed statistical methods used in comparative genomics analysis
Leveraged single cell resolution data to improve understanding of cell type-specific transcriptional responses
Investigated how single-cell RNAseq data and single-cell ATACseq data from mouse hearts correlated with each other across drug treatment and disease states to successfully predict enhancer activation due to heart stress
Established a level of correlation between the two datasets and built a support vector machine (SVM) model to predict enhancer activation (single cell ATAC-seq) from expression data (single cell RNA-seq)
Tools: Python, Scikit-learn, Support Vector Machines, CLI

June 2020 - November 2020 | San Francisco, CA

Projects

                
Music Player Web-App
                  A music streaming web app based on Django
                
AccomplishmentsTools: Django, HTML, CSS, Bootstrap, SQLite, AWS S3, Heroku
Register/login to the web app(with OAuth-based Google Sign-In).
Search and filter songs based on language and singer.
Create multiple playlists and add/remove songs to/from playlist.
Scroll through recently played/viewed songs.

Quiz Web-App
                  A quiz playing web app based on Django
                
AccomplishmentsTools: Django, HTML, CSS, Bootstrap, SQLite, Heroku
Register/login to the web app(with OAuth-based Google Sign-In).
Play Quiz and see the leaderboard

Blog Web-App
                  A simple and extensible blog web-app based on Flask.
                
AccomplishmentsTools: HTML, CSS, Bootstrap, Flask, SQLAlchemy, Postgresql, Python
Users can view posts and contact the admin via Contact Page.
Admin can Add, Delete, Update posts.

Visual Question Answering
                  An attention-based classification model that aims at generating an answer for a given input image.
                
AccomplishmentsIncorporated Convolution Neural Networks (CNN) for extracting image features and Long Short Term
                    Memory for extracting question
                    embeddings.
Tested the model on the COCO dataset, abstract scenes images, and got 69% overall accuracy on the
                    VQA evaluation metric.

Video Summarizer
                  A Seq2Seq model that generates a short summary of the given input video.
                
AccomplishmentsIncorporated CNN to detect and classify objects in the video frames and Long Short Term Memory for
                    generating a summary.
Evaluated the model on MSVD (Microsoft Video Description Corpus) dataset; achieved 0.77, 0.71,
                    0.52 scores respectively on ROGUE,
                    BLEU, METEOR evaluation metrics.

Image Generator
                  An image generator based on the concept of adversarial networks (GANs)
                
AccomplishmentsDeveloped system was tested on a human-face database and loss was calculated by comparing the PCAs
                    of generated and original image.
Calculated difference in PCA was less than 10%, depicting the successful generation of an image by
                    the generator.

Head Counting System
                  A system that calculates the attendance of the class from a panoramic image of a live classroom.
                
AccomplishmentsUsed Singular Value Decomposition for image compression; applied various image processing
                    techniques and morphological operations to
                    detect the number of heads.