Prediction of lithium response using genomic data.

摘要: Predicting lithium response prior to treatment could both expedite therapy and avoid exposure side effects. Since responsiveness may be heritable, its predictability based on genomic data is of interest. We thus evaluate the degree which can predicted with a machine learning (ML) approach using data. Using largest existing dataset in literature (n = 2210 across 14 international sites; 29% responders), we evaluated 47,465 genotyped single nucleotide polymorphisms supervised ML approach. Under appropriate cross-validation procedures, above-chance levels two constituent sites (Halifax, Cohen’s kappa 0.15, 95% confidence interval, CI [0.07, 0.24]; Wurzburg, 0.2 [0.1, 0.3]). Variants shared importance these models showed over-representation postsynaptic membrane related genes. Lithium was not predictable pooled (kappa 0.02 [− 0.01, 0.04]), although non-trivial performance achieved within restricted including only those patients followed prospectively 0.09 [0.04, 0.14]). Genomic classification remains promising but difficult task. Classification potentially improved by further harmonization collection procedures.