This repository is a python implementation of a Poission-emission hidden Markov Model (HMM) to detect mutation hotspots in transcript.
The functions in this repository calculate the posterior probability of a position to be a mutation hotspot and filtered based on expectation.
Functions and required files
Poisson_HMM.py: main function, takes input of mutations counts, run hidden Markov Model and write results to output
hmm.py: function class of Poission-emission hidden Markov Model. For emission, we set the minimum of hidden state mean to be the intial value in each iteration; For transition, we take an average of the Baum-Welsh result with initial value if the transition probality is less than the initial value.
base.py: basic hidden Markov Model functions
MIS_counts.txt: all somatic missense mutations at each amino acid position for each transcipt from COSMIC
COSMIC_counts.txt: cancer mutations counts of different caterogies(missense and silent) denovo_rate.csv: gemline de novo mutation rate of different caterogies(missense and silent) we used observed cancer mutations data and gemline de novo mutation rate to infer the 95% upper limit fraction of missense mutations that are drivers in each gene
To run the programs, numpy and scipy packages are required, output will be in result/hotspot_pos.txt.
Example:
python Poisson_HMM.py