Research Projects

AutoMAxO

Summary:

AutoMAxO is a tool that automates the curation of the Medical Action Ontology (MAxO) database. Using PubMed-BERT and large language models, AutoMAxO extracts disease-treatment relationships from PubMed abstracts and integrates them into the MAxO database. Testing the tool on 21 rare diseases resulted in over 500 novel ontology entries, significantly enhancing the database's scope and usability.

Supervisors:

Prof. Peter N. Robinson, Dr. Justin Reese, Dr. Caufield Harry, Dr. Chriss Mungall

Affiliations:

Trinity College - CT, Jackson Laboratory for Genomic Medicine - CT, Lawrence Berkeley National Lab - CA.

Citation:

Niyonkuru E, Caufield JH, Carmody LC, Gargano MA, Toro S, Whetzel PL, Blau H, Gomez MS, Casiraghi E, Chimirri L, Reese JT, Valentini G, Haendel MA, Mungall CJ, Robinson PN. Leveraging Generative AI to Accelerate Biocuration of Medical Actions for Rare Disease. medRxiv [Preprint]. 2024 Aug 22:2024.08.22.24310814. doi:10.1101/2024.08.22.24310814. PMID: 39228707; PMCID: PMC11370550. [Submitted - DATABASE Journal]. Pre-print: https://www.medrxiv.org/content/10.1101/2024.08.22.24310814v1.

Summary:

WordNet2Vec, a pipeline designed to improve biomedical concept embeddings. Using the WordNet library to standardize terminology, I processed over 30 million PubMed abstracts, achieving an 8% improvement in clustering accuracy for biomedical concepts.

Supervisors:

Prof. Peter N. Robinson, Dr. Hannah Blau

Affiliations:

Trinity College - CT, Jackson Laboratory for Genomic Medicine - CT

Citation:

Niyonkuru E, Gomez MS, Casiraghi E, Antogiovanni S, Blau H, Reese JT, Valentini G, Robinson PN. Replacing non- biomedical concepts improves embedding of biomedical concepts. bioRxiv [Preprint]. 2024 Jul 4:2024.07.01.601556. doi: 10.1101/2024.07.01.601556. PMID: 39005436; PMCID: PMC11244985. [Submitted - PLOS ONE Journal]. Pre-print: https://www.biorxiv.org/content/10.1101/2024.07.01.601556v1.

WordNet2Vec

Summary:

This research presents a CNN-LSTM model to forecast short-term coronal hole activity using image and tabular data from NASA's Solar Dynamics Observatory. The model outperforms traditional methods (LSTM, GRU, ARIMA), reducing errors by up to 34%. This work addresses gaps in predictive modeling, aiding space weather forecasting and preparedness.

Supervisors:

Dr. Chandranil Chakraborttii

Affiliations:

Trinity College - CT, NASA Solar Dynamics Observatory.

Citation:

Solamy, T.; Aritra, B. D. S.; Niyonkuru, E.; Antogiovanni, S.; Chakraborttii, C. Predicting Coronal Hole Activity: Key to Mitigating Space Weather Impacts [Accepted - 2024 Pan-African Artificial Intelligence and Smart Systems Conference].d

Forecasting Coronal Hole Activity with CNN-LSTM Models

"Try to be a rainbow in someone's cloud." ~ Maya Angelou