Big Data and New Models Needed to Study DNA Variation in Evolution and Cancer

on YouTube ↗57m

" summary

Project aims to sequence 10,000 vertebrate genomes and turn tissues into iPS-derived cortical neurospheres, then use RNA‑seq and chromatin assays to pinpoint specific regulatory changes (e.g., prolonged HES5 expression, delayed neuronal maturation) among ~15 million human‑chimp differences. Genomes get modeled as sequence graphs and treated with group‑theory spectral methods; lack of a fast genome rearrangement Fourier transform is an open problem. Cancer results show massive chromothripsis creating circular, amplified oncogene rings; scaling requires ~100 petabytes and open data instead of siloed repositories.

AI-written summary · about

" chapters (12)

1. 0:00Start
2. 0:53The Genome 10K Community Goal: To understand how complex animal life evolved through changes in DNA and use this knowledge to become better stowards of the planet.
3. 2:24Grand scientific challenge of vertebrate molecular evolution Reconstruct the evolutionary history of each base in the genomes of the living species
4. 14:58History graphs add descent edges to sequence graphs
5. 16:30Stochastic Models of Genome Evolution: the Jukes-Cantor model of base substitution
6. 23:44Let the state x, be an arbitrary origin. The stabilizer subgroup H = H, is the subgroup of actions in G that leave x, fixed. For matchings, H is the hyperoctahedral group of symmetries of the n-cube. States in
7. 30:36The Fourier Transform is a linear mapping that converts convolution into multiplication
8. 34:08Thus, if f is taken from a radial rate matrix R (i.e. rate depending only on differences in D) and t is any amount of time, the matrix of probabilities of state changes over various differences is
9. 35:38There is no known computationally tractable closed-form formula for the integers in the Fourier transform matrix for matchings.
10. 39:01Some motivations for large-scale application of comparative genomics in cancer Bring data to research and insights to clinical practice Learn to link phenotypes, including clinical outcomes, to
11. 41:27Different Requirements for 1M Genomes
12. 42:58Extracting molecular state from raw DNA reads

" related talks

• 12mHow data science can make your genome actionableRiley Ennis · Collision 2017
• 28mSome Current Computational Challenges in BioinformaticsRon Shamir · KARPfest80
• 01h 31mIntroduction to Cancer Bioinformatics I: Inferring Genomic Variation from Tumor Sequencing DataBen Raphael, Niko Beerenwinkel · Algorithmic Challenges in Genomics Boot Camp
• 42mPython’s role in accelerating biomedical scienceLi Ting Chen · PyCon APAC 2019
• 14mBig data, AI, the genome, and everythingVijay Narayanan · Strata + Hadoop World 2017
• 43mAlgorithms for Inferring Phenotypes from Ancient DNATony Capra · From Algorithms to Discovery in Genome-Scale Biology and Medicine 2022
• 32mHow machine learning helps cancer researchEvelina Gabasova · Strange Loop 2015
• 50mCancer Genomics - a biologist and a developerLynn Langit, Samantha Langit · NDC Oslo 2017