Periodic Reporting for period 1 - PeptiMOL (Modeling the pharmacokinetics profiles of therapeutic peptides by chemoinformatics methods)
Période du rapport: 2020-05-01 au 2022-04-30
Peptides are defined as macromolecules composed by 2-50 amino acids. Peptides possess multiple therapeutic applications, such as antivirals, antifungals, antibiotics, modulators of the immune, cardiovascular and nervous systems, in addition to their utility in diagnosis.
Peptide drug discovery (PDD) has experienced renewed interest and momentum over the two last decades, thanks to the greater appreciation of the interesting advantages of peptides as an alternative to small molecules: high specificity and activity, easy degradation, do not yield toxic metabolites, and may be reutilized by the organism instead of being converted into waste products.
Nevertheless, therapeutically relevant peptides generally exhibit limited capacity to diffuse across biomembranes such as the human gastrointestinal epithelium, in addition to their low stability. Moreover, due the short plasmatic half-life and low stability of these peptides, they are administered through injections, often several times a day. For those reasons, it is essential to develop methods for modelling the bioactivity of peptides, predict their pharmacokinetic profiles and ultimately allow for the design of novel peptide chains adapted to predetermined bioactivity profiles. Such modelling systems will allow the design of peptides with favourable therapeutic efficacy, and above all, ensure their adequate bioavailability and (preferably oral) administration.
Based on the previous background, the objectives of PeptiMOL were the following:
1. Define parameters (numerical molecular descriptors) to characterize the structural, compositional and physicochemical properties of peptides and develop a user-friendly Java-based tool for their computation.
2. Build mathematical models to predict the PK properties of peptides using the state of the art on statistical and machine learning techniques.
3. Implement the developed models in a Java-based chemoinformatic platform which will enable potential end-users to virtually screen peptide libraries or design novel peptide structures with desirable physicochemical and PK profiles.
Peptide drug discovery (PDD) has experienced renewed interest and momentum over the two last decades, thanks to the greater appreciation of the interesting advantages of peptides as an alternative to small molecules: high specificity and activity, easy degradation, do not yield toxic metabolites, and may be reutilized by the organism instead of being converted into waste products.
Nevertheless, therapeutically relevant peptides generally exhibit limited capacity to diffuse across biomembranes such as the human gastrointestinal epithelium, in addition to their low stability. Moreover, due the short plasmatic half-life and low stability of these peptides, they are administered through injections, often several times a day. For those reasons, it is essential to develop methods for modelling the bioactivity of peptides, predict their pharmacokinetic profiles and ultimately allow for the design of novel peptide chains adapted to predetermined bioactivity profiles. Such modelling systems will allow the design of peptides with favourable therapeutic efficacy, and above all, ensure their adequate bioavailability and (preferably oral) administration.
Based on the previous background, the objectives of PeptiMOL were the following:
1. Define parameters (numerical molecular descriptors) to characterize the structural, compositional and physicochemical properties of peptides and develop a user-friendly Java-based tool for their computation.
2. Build mathematical models to predict the PK properties of peptides using the state of the art on statistical and machine learning techniques.
3. Implement the developed models in a Java-based chemoinformatic platform which will enable potential end-users to virtually screen peptide libraries or design novel peptide structures with desirable physicochemical and PK profiles.
The PeptiMOL project has been executed via six work packages as planned:
- WP1: the relevant peptides to develop the chemoinformatic models were selected. First the available QSPKR models on therapeutic peptides were compiled and analyzed, and secondly the information was filtered and curated by means of advanced data mining techniques in order to define the families of peptides for our study.
- WP2: molecular descriptors specifically tailored for peptides were designed and calculated for the datasets coming from WP1. To do this, a user-friendly software called PeptiDesCalculator was developed in Java language.
- WP3: QSPKR models were developed by applying a variety of statistical and machine learning techniques. Those models were possible thanks to a previous selection of the most adapted descriptors, and a random split of the datasets into training and validation groups. Each model was assessed for its robustness and predictivity by the calculation of the standard statistical parameters, such as accuracy, sensitivity and specificity. Furthermore, the applicability domain of the models was alto established, to elucidate the chemical space for which the models’ predictions are reliable. Finally, as a proof of concept, an experimental validation of one of the models (solubility) was done on a set of 50 peptides thanks to a secondment in the Centro de Investigación Príncipe Felipe (CIPF) de Valencia, as planned in the proposal.
- WP4: the QSPKR models were implemented in a innovative platform, PeptiKinetics, in order to facilitate the use of any person interested in those models, independently of their level of experience on this field.
- WP5: management and monitoring.
- WP6: dissemination within the industrial community and general public, IPR management.
At the end of the project, the main results achieved were the two pieces of software:
a) PeptiDesCalculator software, a standalone software that computes theoretical descriptors for peptide molecules represented as strings of standard amino acid sequences. PeptiDesCalculator computes a total of 48485 molecular descriptors. PeptiDesCalculator is developed in Java programming language (version 9.0.4) and it was designed to run on any UNIX/LINUX or MAC platforms, as well as on microcomputers running Windows 95, 98, ME, 2000 or XP, Vista, 7 and above. It requires Java(TM) 7 Runtime Environment on the target system and also works on X86 and X64 based architecture.
b) PeptiKinetics software, a standalone software that computes the pharmacokinetic properties of peptide molecules to allow for the design of new peptide chains adapted to the therapeutic profile of interest, and with adequate bioavailability. These properties are implemented as mathematical models developed using an array of statistical and machine learning techniques and peptide descriptors computed with the inhouse PeptiDesCalculator repository. The implemented pharmacokinetics properties include: plasma half-life, Caco-2 permeability (Paap), blood brain barrier passage, aqueous solubility, cell penetrating peptide (CPP) uptake and haemolytic activity.
- Programming Language: PeptiKinetics is developed in Java programming language (version 9.0.4) and can thus be run on any operating system that has the Java Virtual Machine (JVM) installed. PeptiKinetics was designed to run on any UNIX/LINUX or MAC platforms, as well as on microcomputers running Windows 95, 98, ME, 2000 or XP, Vista, 7 and above.
- WP1: the relevant peptides to develop the chemoinformatic models were selected. First the available QSPKR models on therapeutic peptides were compiled and analyzed, and secondly the information was filtered and curated by means of advanced data mining techniques in order to define the families of peptides for our study.
- WP2: molecular descriptors specifically tailored for peptides were designed and calculated for the datasets coming from WP1. To do this, a user-friendly software called PeptiDesCalculator was developed in Java language.
- WP3: QSPKR models were developed by applying a variety of statistical and machine learning techniques. Those models were possible thanks to a previous selection of the most adapted descriptors, and a random split of the datasets into training and validation groups. Each model was assessed for its robustness and predictivity by the calculation of the standard statistical parameters, such as accuracy, sensitivity and specificity. Furthermore, the applicability domain of the models was alto established, to elucidate the chemical space for which the models’ predictions are reliable. Finally, as a proof of concept, an experimental validation of one of the models (solubility) was done on a set of 50 peptides thanks to a secondment in the Centro de Investigación Príncipe Felipe (CIPF) de Valencia, as planned in the proposal.
- WP4: the QSPKR models were implemented in a innovative platform, PeptiKinetics, in order to facilitate the use of any person interested in those models, independently of their level of experience on this field.
- WP5: management and monitoring.
- WP6: dissemination within the industrial community and general public, IPR management.
At the end of the project, the main results achieved were the two pieces of software:
a) PeptiDesCalculator software, a standalone software that computes theoretical descriptors for peptide molecules represented as strings of standard amino acid sequences. PeptiDesCalculator computes a total of 48485 molecular descriptors. PeptiDesCalculator is developed in Java programming language (version 9.0.4) and it was designed to run on any UNIX/LINUX or MAC platforms, as well as on microcomputers running Windows 95, 98, ME, 2000 or XP, Vista, 7 and above. It requires Java(TM) 7 Runtime Environment on the target system and also works on X86 and X64 based architecture.
b) PeptiKinetics software, a standalone software that computes the pharmacokinetic properties of peptide molecules to allow for the design of new peptide chains adapted to the therapeutic profile of interest, and with adequate bioavailability. These properties are implemented as mathematical models developed using an array of statistical and machine learning techniques and peptide descriptors computed with the inhouse PeptiDesCalculator repository. The implemented pharmacokinetics properties include: plasma half-life, Caco-2 permeability (Paap), blood brain barrier passage, aqueous solubility, cell penetrating peptide (CPP) uptake and haemolytic activity.
- Programming Language: PeptiKinetics is developed in Java programming language (version 9.0.4) and can thus be run on any operating system that has the Java Virtual Machine (JVM) installed. PeptiKinetics was designed to run on any UNIX/LINUX or MAC platforms, as well as on microcomputers running Windows 95, 98, ME, 2000 or XP, Vista, 7 and above.
PeptiMOL has consisted in an industrial research initiative aimed at the building of computational models for predicting the PK properties of peptides. Those models have been statistically validated, and one of them (solubility) has also been experimentally validated, thus providing to us a good idea about the advantages and potential limitations of their use.
Two innovative chemoinformatic tools have been programed, PeptiDesCalculator and PeptiKinetics, and they will facilitate the design of novel therapeutic peptides with desired PK and bioactivity profiles. These tools also provide functionalities for data curation, direct download of peptides from public repositories (e.g. https://www.rcsb.org/) and a user-friendly GUI to simplify its use by experts and non-experts in modeling.
Additionally, the workflow proposed herein constitutes an important methodological contribution, since until now there were not systematic computational studies in peptides following the same workflow. Specifically, the most innovative methodological aspects of the present project included:
• The definition of absolutely new descriptors specifically customized for characterizing structural, compositional and physicochemical characteristics of peptides.
• The implementation of the first parallelized and user-friendly software tailored for the computation of peptide descriptors, integrating different modules such as variable selection, data preprocessing and similarity analysis.
• The development of completely novel in silico models for the prediction of an array of peptide PK properties.
• The implementation of those in silico models in an innovative platform devoted to the application of those models by people not familiar with computational approaches, but with a high interest into the development of new therapeutic peptides.
Two innovative chemoinformatic tools have been programed, PeptiDesCalculator and PeptiKinetics, and they will facilitate the design of novel therapeutic peptides with desired PK and bioactivity profiles. These tools also provide functionalities for data curation, direct download of peptides from public repositories (e.g. https://www.rcsb.org/) and a user-friendly GUI to simplify its use by experts and non-experts in modeling.
Additionally, the workflow proposed herein constitutes an important methodological contribution, since until now there were not systematic computational studies in peptides following the same workflow. Specifically, the most innovative methodological aspects of the present project included:
• The definition of absolutely new descriptors specifically customized for characterizing structural, compositional and physicochemical characteristics of peptides.
• The implementation of the first parallelized and user-friendly software tailored for the computation of peptide descriptors, integrating different modules such as variable selection, data preprocessing and similarity analysis.
• The development of completely novel in silico models for the prediction of an array of peptide PK properties.
• The implementation of those in silico models in an innovative platform devoted to the application of those models by people not familiar with computational approaches, but with a high interest into the development of new therapeutic peptides.