Run Autoregressive Hidden Markov Model (ARHMM)
The script below creates SLURM jobs for the VIARHMM. Each job created uses a fixed random state and number of latent states. In our analysis, we used 10 different random states and varied the step size from 2 to 10 for each random state. The train and validation models for the various random states and step sizes are stored in the models/all_models folder. We recomend using the pre-trained models, as training new models from scratch can be time-consuming.
(env_name)$ bash scripts/viarhmm.sh
The script below creates SLURM jobs for the time-varying Cox proportional hazard (TVCPH) models. The list of phenotypes included in our analysis are listed in ls_phenotypes.txt file. Along with the phenotype, we also pass metadata information which includes the type of phenotype (mode = prepost or timeint or resilience) and if the phenotype is weighted or not (a boolean value). Some of the phenotypes also account for the uncertainity of the state by using a weighted average approach, where the weights are obtained from the posterior probabilities of the latent state. The results are stored in the data/tvcph folder.
(env_name)$ bash scripts/tvcph.sh
The heritability analysis was done using the Gene x Environment Mixed Model (GxEMM). To run the script, you will need to clone the GxEMM package and ensure that the do_qtl is in the do_bwd folder. The script below creates a SLURM job for each phenotype listed in the gwas_phenotypes.txt and computes its heritability. The results are stored in the data/heritability folder.
(env_name)$ bash scripts/heritability.sh
The GWAS analysis was also done using the GxEMM package. The script creates a SLURM job for each chromsome and phenotype listed in the gwas_phenotypes.txt. The results are stored in the data/gwas folder.
(env_name)$ bash scripts/gwas.sh
After the genetic mapping jobs are completed and the peaks are identified, fine-mapping analysis can be done is a similar way using the GxEMM package. The results are stored in the data/fmap folder. We used the rqtl2 package to generate the fine-mapping plots.
(env_name)$ bash scripts/finemap.sh