Data Science & Machine Learning

Wednesday – October 30th

Session chair: Andras Zsom & Paul Stey

3:10 – 3:20

Session Intro and Welcome

Paul Stey

Director of Data Science and Scientific Computing, CCV

3:20 – 3:50

Identification of Multi-Nucleated Germ Cells, a Marker of Phthalate Toxicity

Prof. Daniel Spade

Assistant Professor, Pathology & Laboratory Medicine, Brown University

Phthalate esters (phthalates) are a class of compounds used to plasticize polyvinyl chloride and as components of various industrial and personal care products. Human exposure to phthalates is nearly universal, and phthalates are known male reproductive toxicants, which creates a concern for human health. Testicular toxicity of phthalates is complex, involving both anti-androgenic effects and impairment of testis morphogenesis. Effects on testis morphogenesis are more difficult and time-consuming to quantify than anti-androgenic effects, and as a result are not factored into risk assessment for phthalates. To address this problem, we sought to develop a method for automatic identification of multinucleated germ cells (MNGs), a marker of phthalate toxicity, in digital images of fetal rat histological sections.

3:50 – 4:20

Biologically Annotated Neural Networks for Multi-Scale Genomic Discovery in Genome-Wide Association Studies

Pinar Demetci

Computational Biology PhD student in the Crawford Lab

Genome-wide association (GWA) studies have provided compelling genetic associations for human complex traits and diseases, contributing to therapeutic target discovery and disease risk prediction. Unfortunately, associations discovered by current GWA methods only explain a modest proportion of the estimated heritability of these traits. Non-linear genetic effects have been proposed as a key contributor to this missing heritability problem.

4:20 – 5:00

Is this a tipping point or am I just biased?

August Guang

Genomics Data Scientist, CCV

The term “tipping point” has become part of everyday English language, used to describe any kind of dramatic shift from which there is no return. However, actually defining and predicting a tipping point is extremely difficult, with the phenomenon often only becoming apparent post-hoc. Nevertheless, organizations make policy decisions based on the consequences of perceived tipping points. In light of this, we sought to understand: what characteristics make an individual more or less likely to declare a tipping point? Using logistic and random forest regression on survey data from 178 undergraduate and graduate students, we find that the perception of tipping points is primarily dependent on the characteristics of the graph, and secondarily on their own experience or emotions about tipping points. These conclusions have implications for management and sensemaking of perceived tipping points.