Skip to content

A deep learning system accurately classifies primary and metastatic cancers using passenger mutation patterns #
Find similar titles

2,606명 WGS에서 추출한 Somatic passenger mutation으로 24개의 암종을 예측함.

암환자의 WGS 자료에서 somatic mutation의 분포와 종류를 학습하여, 24개의 암종을 91% 정확도로 예측함. 의외의 재밌는 점은 driver 유전자 정보를 추가했을 때 정확도가 낮아짐. 그로 인해, 제목에 passenger mutation 이라 명시

WGS 자료에서 다음의 특징들을 따로 추출하여 모델 학습에 사용하였다.

  • Mutation distribution
    • SNV-BIN (2897): Number of SNVs per 1-Mbp bin and per chromosome, normalised against the total number of SNVs per sample
    • CNA-BIN (2826): Number of CNAs per 1-Mbp bin
    • SV-BIN (2929): Number of SVs per 1-Mbp bin, and per chromosome, normalised against the total number of SV per sample
    • INDEL-BIN (2757): Number of SNVs per 1-Mbp bin, and per chromosome, normalised against the total number of INDEL per sample
  • Mutation type
    • MUT-WGS (150): Type of single-nucleotide substitution, double- and triple-nucleotide substitution (plus its adjacent nucleotide neighbours)
  • Driver gene/pathway
    • GEN (554): Presence of an impactful mutation in a suspected driver gene
    • MOD (1865): Presence of an impactful mutation in a gene belonging to a suspected driver pathway

Suggested Pages #

web biohackers.net
0.0.1_20140628_0