Genetic and Psychosocial Predictors of Aggression: Variable Selection and Model Building With Component-Wise Gradient...

Area of Science:

• Behavioral Science
• Genetics
• Data Science

Background:

• Machine learning (ML) offers efficient tools for analyzing complex datasets in aggression research.
• ML techniques have not been widely applied to aggression prediction, despite their potential.
• Genetic factors, like FKBP5 polymorphisms, may influence aggressive behavior.

Purpose of the Study:

• To examine predictors of aggression using ML techniques.
• To construct an optimized and parsimonious model for predicting aggressive behavior.
• To investigate demographic, psychometric, and genetic (FKBP5) predictors.

Main Methods:

• Component-wise gradient boosting was used to select salient predictors from an initial set of 20.
• Model reduction via backward elimination was applied to enhance parsimony and generalizability.
• The Buss-Perry Aggression Questionnaire (BPAQ) was used to measure trait aggression.

Main Results:

• Gradient boosting identified 8 key predictors of aggression (R² = 0.66) from 20 variables.
• Backward elimination simplified the model to 6 predictors: smoking status, psychopathy, childhood trauma, and FKBP5_13 gene variant (rs1360780).
• The reduced model retained 99.4% of the initial model's predictive power.

Conclusions:

• ML successfully identified established aggression predictors (psychopathy, trauma) and novel genetic associations (FKBP5).
• The study demonstrates the utility of inductive data science for aggression prediction in large datasets.
• Replication with larger sample sizes is needed to confirm findings, particularly the role of FKBP5.