An approach to identifying aggression profiles via machine learning

The complex nature of aggression and challenges in phenotype description often make it difficult to identify genetic variations associated with aggressive behavior. The use of machine learning algorithms and multivariate analysis makes it possible to construct more accurate diagnostic schemes to characterize the underlying causes of aggression. The aim of this study was to develop a method to differentiate aggression profiles among subjects in a heterogeneous sample and to facilitate the association of the obtained profiles with genetic variations of BDNF rs6265, BDNF rs10835210, HTR2A rs6313, and DRD4 1800955 later.

Participants' aggressiveness was assessed using the Buss–Durkee Hostility Inventory (BDHI). On the basis of the 10 scales of the BDHI, 1013 unique subspaces were formed, in which data were clustered. "Cluster neighborhood" plots were constructed for each participant in the study, allowing us to identify the content of each category of subjects in all of the clusters into which the participant fell. This approach demonstrated two dramatically different phenotypes based on all the possible combinations of characteristics. However, many study participants exhibited mixed phenotypes, which were separated using the DBSCAN spatial clustering algorithm.

Individuals initially classified into a control group, addiction-prone group, or inmate group (convicted of felony) were often adjacent to each other in the newly formed clusters, suggesting the need to reevaluate the sample: 13 individuals from the addiction-prone group and 17 individuals from the inmate group were assigned to the control group, and 24 individuals from the inmate group and 6 individuals from the control group were assigned to the addiction-prone group. A total of 140 individuals belonged to the monotype group, while the remaining 261 individuals belonged to the mixed phenotype group.

Phenotypes defined by indirect traits often create an erroneous picture. It is necessary to identify similar groups of patients based on their psychometric data and then examine their genotypes within these groups.