An application of collaborative targeted maximum likelihood estimation in causal inference and genomics

Area of Science:

• Biostatistics
• Genomics
• Causal Inference

Background:

• Estimating causal effects in genomic data is complex.
• Existing methods like propensity score matching have limitations.
• A novel collaborative double-robust targeted likelihood estimator (C-TMLE) is proposed.

Purpose of the Study:

• To present and apply the C-TMLE for non-parametric estimation of causal effects and variable importance.
• To evaluate C-TMLE's performance against existing estimators.
• To estimate the effect of HIV mutations on lopinavir resistance.

Main Methods:

• Application of the C-TMLE to a point treatment data structure.
• Comparative simulations with augmented inverse probability of treatment weighted (AIPTW) estimators and propensity score methods.
• Utilizing the influence curve for asymptotically valid statistical inference.

Main Results:

• C-TMLE demonstrated superior performance compared to AIPTW and standard inverse probability of treatment weighting.
• The method accurately estimated covariate-adjusted marginal effects of individual HIV mutations on lopinavir resistance.
• Statistically significant mutations identified by C-TMLE strongly agreed with the Stanford HIVdb database.

Conclusions:

• C-TMLE provides a robust and accurate approach for causal inference in genomic studies.
• The estimator is effective for identifying key genetic variants, such as HIV mutations, influencing treatment outcomes.
• This method facilitates reliable statistical inference in complex biological data.