Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Moliere Project Summary Poster (July 2018)


Published on

Presented at the Google Ph.D Intern Research Conference (PIRC 2018).
This poster focuses on the work originally presented in "Moliere, Automatic Biomedical Hypothesis Generation System" including recent in-submission results since.

Published in: Science
  • Be the first to comment

  • Be the first to like this

Moliere Project Summary Poster (July 2018)

  1. 1. 1 1 1 2 2 2 a c c 1 2 Automatic Biomedical Hypothesis Generation Ilya SafroMichael Shtutman HTR[12]A & Venlafaxine Justin Sybrandt Hypothesis Generation Large-Scale Validation - Understand all available research - Identify missed implicit connections - Provide early information to decision makers Fish Oil Raynaud's Syndrome Blood Viscosity - Evaluate thousands of potential hypotheses Generating Hypotheses Automatically Legend: Query Term Path Node Nearby Abstract a c Problem - Construct network with NLM sources - Describe potential connections through shortest-paths - Generate topic model from cloud and apply metrics Abstracts Key n-grams Codified Terms Predicates Embedding Nearest-Neighbors TF-IDF Semantic Medical Database Unified Medical Language System ? - Rediscover exsiting HRT1A connection - Detect lack of HTR2A connection - Uses pre-2010 data HTR1A HTR2A Venlafaxine Depression Anxiety - Initial validation example See more here Ranking Hypotheses - Quantify connections through heuristics - Derived via embeddings & topic network - Example Metrics: Published Connection High Connectivity Noise Connection Low Connectivity Nucleic Acid Bacterium Eukaryote Cell - Distance between terms - Distance to topic centroids - Correlation w.r.t. topic model - Betweenness of topic network - Topic network clustering coef. - Polynomial combination - Evaluate system without expert input - Holdout experiment with historical data A K P C A A A A K K K C KC CP P C Topic from a - ci Keyword i Discovering New Treatments - 25-33% of HIV+ patients develop HIV-Associated Dementia - Apply MOLIERE to identify treatment options - Rank all human genes & create candidate set - Discover treatment option via lab experiments Gene Gene Gene Dementia 40,000 Human Genes HIV - Rank via above metrics ranking lab experiments Pfizer Ends Hunt for Drugs to Treat Alzheimer’s and Parkinson’s About 300 layoffs to take place after once-promising compounds failed during testing WSJ - Medical research needs early investment - 2-4k bio papers published daily - Missed connections lead to missed treatment options, costing lives