Mutations in cancers

• somatic mutations are a major cause of tumorigenesis
  • but mutations are noisy
• two source of recurrent mutations:
  • Background mutation rate
    • mutagens, DNA context, methylation etc
    • no selection pressure
  • Functional selection pressure
    • selected!

background mutation Model is needed

Plan

1. Modelling Background mutation rate
   • Expected: model BMR -> significance of each mutation
2. Analysis functional selected mutations
   • Missense / Nonsense comparison
   • Functional analysis

Model 1

• developed by (Chang et al. 2016) in nbt.
  • trinucleotide mutation rate -> weight-averaged for codon
    
  • [Not] cancer specific
    
  • data process :
    • using all single nucleotide variants(SNVs)
    • removed hypermutated sites: positions bearing >=99% mutations in a gene
      
• binomial model

$$Prob(X=k)=\binom{n}{r}P_{c,g}^{k}(1-P_{c,g})^{n-k} \\ $$

center nucleotide mutation rate in a trinucleotide(t)
$$P_t=\frac{C_t}{F_t} \\\text{ }C_t:\#of\_mut;F_t:\#of\_trinucleotide$$

codon mutation rate $P_{c,g}$ of codon c and gene g:
$$P_{c,g}=\frac{\frac{C_g}{N_{sample}}}{\sum_{t \in g}{N_{t,g}P_t}} \sum_{t \in c}{\frac{n_{t,c}}{n_c}P_t}$$

$N_{t,g}$ : #of tri in gene;  
$N_{sample}$ : sample size ;  
$n_{t,c}$ : # of mut in site t of codon c;  
$n_c$ : # of mut in codon c  

Preliminary results

• data: UCSC (238 samples, 68197 SNV mutations)

Model 2(to do)

• From MutsigCV by (Lawrence et al. 2013)
  • designed for genes
    • might require some tuning-up
  • use silent mutations from target gene and its neighbors

next

• Modelling BMR
  • finish model 1 in complete dataset
  • Combine features from experiment
• Reports to finish.