Computational Challenges in a Densely Sequenced Tree of Life

on YouTube ↗44m

" summary

Genome alignment to huge reference collections breaks: simulations show up to 50% of reads cross-map to a close neighbor, degrading SNV recall especially for alleles absent from the reference. Standard MapQ filtering (mapQ≈10) reduces false positives but slashes recall. Built an exact unique k-mer probe database that genotypes known SNVs ~100× faster than alignment, uses about 25–40 GB RAM so it runs on a laptop, and is more accurate for known sites. Also compresses redundant sets (e.g., 3,646 genomes → 46 tag genomes) to speed SNV discovery.

AI-written summary · about

" chapters (15)

1. 0:00Microbiome Precision Medicine
2. 0:50A very large-scale ‘omics problem
3. 3:49Microbiome Science
4. 4:47Metagenotyping single nucleotide variants (SNV)
5. 7:22Most species had no genome
6. 9:39Genome explosion
7. 11:23Closely related species are common
8. 11:58Read competition in dense lineages
9. 15:14MIDAS2 mitigates low alignment uniqueness & cross-mapping
10. 17:44GTPRO: metagenotyping known SNVs with by exact matching to unique k-mers
11. 20:20GT-PRO strategy works for metagenomes, genomes, contigs, unassembled reads
12. 21:08GTPRO: 100x faster, more accurate SERVER
13. 25:10Genome redundancy offers solution
14. 25:27Maast: fast variant discovery from genomes
15. 27:56Future Prospects