IFPTML AlgorithmsFrom Cheminformatics Models to Software Development, StartupCreation, and Innovation Transference

  1. Bediaga Bañeres, Harbil
Supervised by:
  1. A. Pazos Director
  2. Humberto González Díaz Co-director

Defence university: Universidade da Coruña

Fecha de defensa: 15 December 2023

Committee:
  1. Virginia Mato-Abad Chair
  2. Enrique Onieva Caracuel Secretary
  3. Juan Manuel Ruso Beiras Committee member
Department:
  1. Computer Science and Information Technologies

Type: Thesis

Teseo: 829049 DIALNET lock_openRUC editor

Abstract

The irruption in scence of Drug High Throughput Screening (HTS) technologies has prompted an explosion in the report of pre-clinical assays data for new hit-to-lead compounds with potential as Active Pharmaceutical Ingredients (APIs) in Pharmaceutical Industry. The análisis of all this data with techniques Artificial Intelligence (AI) may lead to the development of new predictive models. These models may be used in turn to predict more specific and safer compounds which the consequent reduction of costs in time and resourcers in APIs development. However, AI analysis of this presents many of the challenges of Big Data problems. It means, shortly, data analysis problems with issues related to Volumen, Velocity, Veracity, Variability, Value, and Complexity(5V + C). The first and second Vs are more or less self-explained and the Variability, Veracity, Value, and Complexity issues refers to data with problems of missing data, not consistent tendencies, errors, contradictory reports, interrelations such as co-llinearity/codependent labels forming complex networks, read-across (multi-species, multi-ouput, multiscale) information, perturbations in multiple input/output variables, multi-labelling problems, etc. In this context, our group introduced the Information Fusion, Perturbation Theory, and Machine Learning (IFPTML) algorithm to facilitate the development of read-across multi-output models able to predict multiple outcomes of chemical compounds/drugs in pre-clinical assays. Our group has also reported a software calle SOFT.PTML which is a general puropose platform for IFPTML modeling. However, many aspects are yet to be covered. Many problems such as anti-cancer compounds discovery, allosteric compounds assays studies have not been analyzed with IFPTML algorithms. In addition, specific user-friendly software for these problemas has not been already reported. Last, despite the potential application in industry no startuptup company was developed (at the beginning of this tesis) for the transference of IFPTML technology to industry. Consequently, the objective of this tesis focus firstly on the development of new IFPTML models of anti-cancer compounds and allosteric compounds assays. Next, we report the development of the user-friendly software LAGA for the read-across prediction of anti-cancer compounds. Last, we describe the planning, creation, structure, services, etc. of IKERDATA S.L a new inter-university startup company focused on the transference of IFPTML technology to Galician and Basque Country companies in the first instance with perspectives of Spain, Europe, and Worldwide proyection.