Fitness effects of HIV mutations Lucy Crooks Theoretical Biology, ETH Zurich

Drug-resistance in HIV Combination therapy (3 drugs) most effective if resistance mutations have negative effects under some drugs How mutations interact to affect fitness also has an influence through recombination negative epistasis can accelerate evolution epistasis is also relevant for theories of the evolution of sexual reproduction

Aims estimate the fitness effects of HIV-1 mutations estimate interactions between these effects 17,000 sequences + fitness in 16 treatments

Data 400 positions 1,800 mutations 180,000 pairwise interactions complex mutational patterns

My Approach randomly split interaction terms into subsets fit a series of models each with main effects for all mutations remove terms with high p-values (t-test of coefficient) repeat until few enough interactions to fit into one model GLM with variance  mean p-value cut-off = 0.4 significance tested by change in deviance (p>0.05)

Approach (2) fit remaining terms into one model sequentially remove sets of terms with highest p-values repeat until only significant terms remain (p>0.05)

Technical details each model run as separate job fitting done in R with model matrix generated in perl method is iterative weighted least squares using QR decomposition (calls fortran routine dqrls) 1 processor, exclusive node use CPU time = 5 hours

Preliminary results fitness effects of mutations in absence of drugs

Preliminary results (2) epistasis in the absence of drugs

Outlook simplify the model test robustness of subset approach repeat analysis for 15 drug treatments find funding!