Machine learning team from Jefferson’s Computational Medicine Center (CMC) reaches top ranks among over 50 teams in latest “DREAM Single Cell Transcriptomics Challenge”
For more than a decade, the DREAM initiative has driven crowd-sourced open science scientific contests in different areas of biology and medicine. Nestoras Karathanasis and Phillipe Loher represented Jefferson’s CMC in the DREAM Single Cell Transcriptomics Challenge that ended in early December 2018.
Single-cell transcriptomics sequencing allows one to gain knowledge about the abundance of protein-coding and non-coding genes in only a few thousand cells from a given tissue. Such an approach does not suffer from the drawback of standard approaches where the requirements for a large starting population of cells can obscure the existence of subgroups among them with distinct identities. Single cell approaches do not maintain the spatial arrangement of the cells. This latest DREAM challenge used Drosophila embryo as a model system, and sought to determine whether one can a) tackle the spatial arrangement problem by devising novel methods, and, b) identify those genes that carry the most spatial information.
The CMC team’s approach leveraged machine learning techniques and big data approaches. They were able to correctly establish a cell’s spatial arrangement as well as identify de novo many of the genes that were known (to the challenge’s organizers) to carry the most spatial information. Importantly, the team identified many novel genes that the analysis shows carry important positional information and were not previously suspected. The team’s approach included dimensionality reduction through an innovative feature selection strategy while the algorithms were implemented using public libraries that included LASSO, Deep Neural Networks, and PyTorch.
In each of the three sub-challenges, the team placed in the top–10 among more than 50 teams from all over the world. Specifically, they ranked 10th, 6th, and 4th in sub-challenges 1, 2 and 3, respectively. As part of their participation in the DREAM challenge, the team prepared a detailed document that contains a step-by-step guide to their methods as well as released publicly their source codes. The CMC team has also been invited to work on a manuscript alongside the challenge organizers in highlighting their contribution.
Congratulations to Nestoras and Phillipe on their initiative, great performance, and their continuing efforts to break new ground and contribute new knowledge to the field of biology.