• machine learning
• virtual screening
• cdk9

The purpose of this project is the development of a Machine Learning platform for the identification of new promising inhibitors of the cyclin-dipendent kinase 9 (CDK9). The CDK9 protein plays a crucial role in the pre-mRNA transcription regulation, as it influences both the capability of the RNA polymerase II to successfully escape from the Early Elongation Checkpoint and the termination phase of the transcriptional process. The abnormal activity of the CDK9 is involved in a plethora of illnesses, many of which are tumoral diseases: this results in CDK9 being a target of great interest in the scientific field, and in the discovery of new CDK9 inhibitors being an important need to be met. With this purpose, we have developed 70 Machine Learning models, each of them created with the aim of being capable of discerning active from inactive compounds on this target. The analysis of models’ performances, evaluated both individually and in combination according to different approaches, has allowed us to find the most suitable asset for developing the platform previously mentioned. The Machine Learning platform developed was then used in a Virtual Screening protocol applied to over 7 millions compounds, leading to the identification of promising CDK9 inhibitors as a result. Furthermore, the evaluation of the models previously conducted has allowed us to set up another investigation with a different purpose: indentifying promising CDK9 inhibitors among compounds already known for their activity against not-kinase targets. In this analysis a new Machine Learning platform has been developed by choosing the most appropriate model for this purpose within those created: this platform was used for screening compounds tested on not-kinases targets retrieved from CHEMBL35 database. Molecules resulted from this analysis represent new possible CDK9 inhibitors with previously unknown off-target activity.