Through the use of proprietary software, statistical analysis tools and machine learning algorithms, backed by substantial computer hardware facilities, First Health Pharmaceuticals brought computational drug development to an entirely new level



The First Health Pharmaceuticals’ approach is a connubium of both “classical” and cutting edge Computer Aided Drug Design techniques and algorithms, which makes it possible to minimize the attrition rate of our drug discovery programs.

Classical techniques such as pharmacophore modeling and docking, which have a proven back track of successful applications, are integrated with proprietary statistical models based on Machine Learning algorithms, and modern protocols that accurately estimate the free energy of ligand / enzyme binding. The huge computational workflow involved in all of this is supported by a powerful water cooled hardware infrastructure with a total of 44 Teraflops and no less than 216 CPU core threads of calculating power.


In the last decade, the creation of publicly available data repositories such as Pubchem and ChEMBL as well as screening initiatives like the NCI Developmental Therapeutics Program (DTP) made a huge amount of drug discovery data available to the scientific community. To fruitfully mine this vast amount of information, both experienced researcher and advanced machine learning algorithms are required.


The main objective of First Health Pharmaceuticals is the identification of human RNA Helicases Inhibitors involved in both cancer pathogenesis and virus infection. By looking at the ADME (an acronym of Absorption, Distribution, Metabolism and Excretion) and physical-chemical properties from the first stages of the Research and Development program, the research efforts are focused on the most promising compounds from the very start. The application of the aforementioned approach significantly accelerates the identification of good clinical candidates.


First Health Pharmaceuticals employs the following advanced computational approaches

Modeling of physical-chemical properties: Machine learning models (Random Forrest, Artificial Neural Network, Bayesian classifier).

Co-/Off-target identification: Reverse 3D-Pharmacophore screening of Protein Data Bank protein/ligand complexes, Polypharmacology profile by Tanimoto similarity search in PubChem/ChEMBL.

Drug re-purposing: Screening of known drugs by 3D Pharmacophore/Docking.

In silico study/understanding of protein inhibitor interaction: Metadynamics/Alchemical free energy perturbation.