I am a computer science PhD Student in MIT's Computer Science and Artificial Intelligence Laboratory and Brigham and Women's Hospital's Divison of Genetics and Center for Data Sciences where I am lucky enough to be co-advised by Bonnie Berger (MIT) and Po-Ru Loh (Harvard). With the rise of high-throughput DNA sequencing and electronic health care records, the medical field is producing more data than nearly any other human endeavour. Buried in the billions of bytes of data produced every day are deep insights into the origins of and improved treatments for complex diseases. The challenge is to efficiently leverage meaning from the tidal wave of data. I design new stastistical and machine learning algorithms to meet this challenge. My ultimate goal is to enable medicine to provide personalized care tailored to every person's individual qualities and needs.
I also think a good deal about better ways to help people understand probability, statistics, and machine learning. I am particularly interested in ways to help laypeople gain an intuitive understanding of the abstract concepts which underly the probabilistic models they interact with everyday; to this end, I maintain a small (but growing) github repository of demonstrations I hope others might find useful.
PhD Computer Science MITCambridge, MA 2018-Present
MA Mathematics University of CambridgeCambridge, UK 2014-2015
Awarded with Distinction (Highest Honors)
BS Applied Mathematics & Biology Brown UniversityProvidence, RI 2010-2014
Magna cum Laude with Honors
Mosaic Copy number Variation in Autism
Mosaic copy number variation is a rare genetic disorder in which some fraction of an individual's cells carry too much or too little DNA. Such mutations arise spontaneously during fetal development. In collaboration with Boston Children's Hospital and Harvard Medical School I am developing an algorithm to detect this mutation using inexpensive consumer genetic tests. In preliminary analysis we have found that up to 1% of autism cases may be affected by large, deleterious mosaic CNVs. Our approach could help identify hundreds of children at increased risk of autism each year at minimal cost using technologies already available to clinicians.
Collaborators: Rachel Rodin, Christopher Walsh, and Peter Park
Deep Insights into Cancer Genomics
The central challenge in cancer biology is to understand how and why specific mutations cause cancer. This challenge is monumentally difficult because 1) every type of cancer is different (e.g. liver cancer carries different mutations than skin cancer) and 2) tumors carry thousands of upon thousands of mutations, but the vast majority of these are harmless passengers which hide the handful of truly pathogenic driver mutations. Using recurrent neural networks to learn tissue-specific tumor features, my collaborator Adam Yaari and I are modeling the background mutation rate of individual cancers with unprecedented resolution. In our pilot test, we modeled the mutations of skin melanoma with 100-fold better resolution than other current state-of-the-are methods. We are now using these results to identify novel driver events.
Collaborators: Adam Yaari
Neuron-specific regulators of neurological disorders
Over the last decade, genome-wide association studies have been tremendously successful at finding genomic regions associated with diseases. However, our understanding of these relationships is limited by the age old correlation vs. causation problem. Working with collaborators in the Harvard's Department of Neuroscience, I am adapting a method known as LD Score Regression created by Hilary Finucane in order to investigate which specific DNA regulatory elements in which specific neuron subtypes contributed most to the genetic architecture of neuropsychiatic disorders such as schizophrenia and autism.
Collaborators: Gabriella Boulting, Bulent Ataman, Michael Greenberg
Detecting somatic mutations in brain tissue
As part of the Brain Somatic Mosaicism Consortium, I am involved in determining and setting gold-standard practices for detecting somatic mutations in the brain. Specifically, I head up efforts to develop statistical methods which can pool evidence across multiple sequencing modalities to gain power to detect these rare, low-signal events.
Collaborators: Taejeong Bae, Yifan Wang, Sean Cho, Alexej Abyzov, Jeffrey Kidd, Alex Urban
Somatic mutations accumulate in neurons with age
The DNA in our cells are bombarded every second of the day. This relentless assault occasionally results in mutations. In actively dividing cells, these mutations can result in cancer. However, the vast majority of the cells in the human body do not divide. What happens as these cells acquire mutations? I was part of a collaborative team which answered this question in neurons. We found mutations slowly but surely build up in neurons during healthy again. Intriguingly, the somatic mutations accumulate much more rapidly in diseases which cause early-onset aging (so-called progeroid diseases).
Quality control for single-cell whole-genome sequencing
A powerful tool to understand genetic heterogeneity is single-cell whole genome sequencing in which the entire genome of a single cell is sequenced. Unfortunately, the technology required to perform this minor miracle creates significant bias and artifacts in the data. If left uncorrected, these biases seriously confound results. Adapting methods from signal processing, I developed a suite of tools to characterize these biases and help enable their proper removal.
Determining the origin of the neocortical beta wave
Even when at rest, our brains are constantly active. This activity most often takes the form of oscillating rhythms. The beta wave (15-30 Hz rhythmic activity) is one of the dominant signals observed in the human brain. However, its origin remained a mystery for nearly 100 years. Using a combination of biophysically realistic modeling, non-invasive brain recordings in humans, and in-vivo recordings from mice and monkeys, I predicted and validated the first known in-vivo origin of this rhythm.
Sports: Alpine skiing, water skiing, sailing, scuba diving, hiking.
Other: Cooking for larger dinner parties.
© Maxwell Sherman 2019