Multivariate Pattern Analysis of Genotype-Phenotype Relationships in Schizophrenia

Area of Science:

• Psychiatric Genetics
• Computational Neuroscience

Background:

• Schizophrenia exhibits significant clinical and etiological heterogeneity.
• Genetic risk variants are linked to various schizophrenia phenotypes.
• Current methods like polygenic risk scores have limitations in capturing complex genetic effects.

Purpose of the Study:

• To compare the predictive power of random forest algorithms versus traditional polygenic risk scores for cognitive endophenotypes in schizophrenia.
• To investigate the utility of flexible machine learning algorithms in genotype-phenotype mapping for complex psychiatric disorders.

Main Methods:

• Utilized a combined sample (N=739) and an independent validation sample (N=336).
• Employed random forest and linear regression/polygenic risk scores to predict 6 cognitive endophenotypes using 77 schizophrenia-associated single nucleotide polymorphisms (SNPs).
• Applied tenfold cross-validation for model evaluation in the discovery sample.

Main Results:

• Random forest models accurately predicted processing speed and visual memory.
• Polygenic risk scores successfully predicted verbal and working memory.
• Both approaches demonstrated generalization to an independent sample with comparable predictive performance.

Conclusions:

• Both random forest and polygenic risk scores are valuable for mapping specific genetic variants to schizophrenia-related phenotypes.
• Flexible algorithms like random forest offer potential for a more comprehensive analysis of genotype-phenotype relationships in schizophrenia.
• These findings contribute to parsing the heterogeneity within schizophrenia through advanced genetic analysis.