Use of a machine learning approach to estimate pathobiological effects of crack cocaine administration in rats

This study aimed to apply a machine learning workflow to identify the most relevant biological parameters for predicting both the dose and the route of crack cocaine administration.Seventeen variables were evaluated in rats exposed to different doses of crack cocaine, either intraperitoneally (18 or 36 mg/kg, i.p.) or via passive inhalation (25, 50, or 100 mg). Random forest (RF) analysis was used to build predictive models, feature importance analysis to identify key variables, and interaction dependence analysis to explore relationships among variables.Eighty percent of the data was used for training and 20% for testing. The model achieved 85% accuracy in the training phase and 100% in the test phase. During training, the highest accuracy was observed for the 100 mg inhaled group, while the lowest was for the 50 mg inhaled group. Notably, 20% of the 50 mg i.p. cases were misclassified as 36 mg i.p. Feature importance analysis highlighted four key predictors: liver karyolysis, kidney Ki-67 expression, liver binucleation, and escape behaviour.These findings demonstrate that machine learning (ML) can accurately predict both dose and route of crack cocaine exposure, and can highlight biologically relevant parameters involved in the drug’s systemic and behavioural effects. This study aimed to apply a machine learning workflow to identify the most relevant biological parameters for predicting both the dose and the route of crack cocaine administration. Seventeen variables were evaluated in rats exposed to different doses of crack cocaine, either intraperitoneally (18 or 36 mg/kg, i.p.) or via passive inhalation (25, 50, or 100 mg). Random forest (RF) analysis was used to build predictive models, feature importance analysis to identify key variables, and interaction dependence analysis to explore relationships among variables. Eighty percent of the data was used for training and 20% for testing. The model achieved 85% accuracy in the training phase and 100% in the test phase. During training, the highest accuracy was observed for the 100 mg inhaled group, while the lowest was for the 50 mg inhaled group. Notably, 20% of the 50 mg i.p. cases were misclassified as 36 mg i.p. Feature importance analysis highlighted four key predictors: liver karyolysis, kidney Ki-67 expression, liver binucleation, and escape behaviour. These findings demonstrate that machine learning (ML) can accurately predict both dose and route of crack cocaine exposure, and can highlight biologically relevant parameters involved in the drug’s systemic and behavioural effects.