AxDrug: AI-Powered Drug Discovery Platform – Accelerate Your Pipeline with PozeSCAF
PozeSCAF’s AxDrug Platform embodies a groundbreaking approach to drug discovery, integrating advanced AI and computational chemistry methodologies to propel Hit Identification, Hit to Lead, and Lead Optimization processes. Leveraging deep learning, machine-learning models, and accelerated computation methods, AxDrug navigates the intricate landscape from target validation to lead optimization with precision and efficacy.
Understanding AxDrug’s Components
1. ChemBio-SAR Module:
The ChemBio-SAR module within AxDrug is a multifaceted engine that harnesses the power of expansive chemical libraries, structure-based generative chemistry, and highly validated predictive models. By employing deep learning and machine learning methods, ChemBio-SAR generates predictive models and seamlessly integrates computational methods to identify druggable pockets within targets. These predictive models also aid in molecule generation, validation for drug-likeness and selectivity, and prioritization of high-affinity compounds. Through these capabilities, AxDrug optimizes crucial properties in early drug discovery, including efficacy, selectivity, and drug-likeness.
Expanding further, we can delve into specific examples of how ChemBio-SAR’s predictive models have been instrumental in identifying promising drug candidates. By analyzing case studies or real-world applications, we can illustrate the tangible impact of AxDrug on accelerating the drug discovery process.
1.1. Chemical Toolbox:
Central to the success of drug discovery endeavors is the selection of optimal candidates from screening libraries. PozeSCAF’s Expansive Chemical Space (ECS), comprising synthetically feasible and drug-like molecules with diverse chemical and shape space, offers a significant advantage in identifying superior compounds. By virtual screening ultra-large libraries, AxDrug enhances the efficiency of drug discovery, minimizing attrition rates in clinical trials and improving potency and physicochemical properties.
Expansive Chemical Space (ECS): PozeSCAF has curated a vast repository of 20 billion compounds from ultra-large chemical databases, enriched further by generating 100 million bio-like compounds using unique generative chemistry algorithms. Enhancing the drug-likeness of compounds within the Expansive Chemical Space (ECS) involves applying various filters including the Rule of Five, Rapid Elimination of Swill (REOS), Pan-Assay Interference Compounds (PAINS), aggregators, and Veber filters. This process ensures the refinement of the compound library to optimize its pharmacological potential.
The compound library is categorized into several databases: The Central Nervous System Active Database (CNSDB), which evaluates compounds based on Blood-Brain Barrier (BBB) scores; the Fragment Database (FrDB), which categorizes compounds by size; the Covalent Compound Database (CCDB), which identifies compounds with reactive warheads; and the Protein-Protein Interaction Database (PPIDB), which assesses compounds based on their properties.
Through our virtual screening protocols, the ECS serves as a critical tool in identifying promising drug-like candidates, thus advancing drug discovery efforts.
1.2. Biological Models/Bio Box:
We evaluate over 10 million compounds against 20,000 biological targets encompassing mammals, bacteria, viruses, and fungi, with only a select few advancing to drug status. These compounds exhibit binding affinity towards therapeutic targets, demonstrate activity against a wide array of diseases, possess pharmacokinetic properties, and reveal various toxicities such as adverse effects, mutagenicity, and carcinogenicity.
The integration of chemography spanning broad chemical space with the vast biological landscape through knowledge hypergraphs (KHG) unveils intricate relationships. This connection facilitates the generation of predictive models using deep neural networks to forecast the biological properties of novel compounds. Within AxDrug’s Biological Models or Bio-box, Knowledge Hypergraphs (KHN) link chemotypes to targets, diseases, pathways, and toxicities or side-effects.
Consequently, biological models expedite every stage of drug discovery, including target identification, validation, lead identification and optimization, polypharmacology, assessment of off-target effects, drug disposition, and toxicities.
To construct the Knowledge Hypergraphs (KHN) and models, we compiled bioactive chemical data and their biological properties from various sources including ChEMBL, bindingDB, drug databases, PharmGKB, chembank, BRENDA, SIDER, NIAID ChemDB, HMDB, PDB, and Uniprot, among others. The chemical space was categorized based on families of binding targets, including GPCRs, Nuclear Hormone Receptors (NHRs), Ion Channels, Kinases, Transporters, Cytokine Receptors, Receptor Tyrosine Phosphatases, Tumor Necrosis Factor Receptors, Enzymes, Transcription Factors, Immune Checkpoint Proteins, Bromodomains, DNA, and RNA. Additionally, the chemical space was segmented into categories such as Anti-infectives, Antimalarials, and Antivirals.
By employing chemographic methods, we mapped the chemical space to targets, targets to diseases, diseases to pathways, targets to pathways, and chemical space to off-target effects, thereby generating hypergraph learning models. Consequently, for any given compound, these models offer insights into potential targets/off-targets it can bind and diseases it can potentially treat.
Through the utilization of hypergraph models, we organized our Expansive Chemical Space(ECS) into smart databases based on target family and disease-centric criteria. These smart databases include kinase-focused libraries, GPCR-specific libraries, and others. These smart libraries expedite the process of rapid virtual screening, circumventing time-consuming computational methods, thereby facilitating the identification of leads.
1.3. AI-Enabled ADMET Models/ADMET tool box
Pharmacokinetic properties influence the efficacy and safety of the drugs. They are crucial for understanding the behavior of a drug within the human body. Predicting absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties is essential in drug discovery and development to identify potential candidates with favourable pharmacokinetic profiles and reduce the risk of adverse effects. Early assessment of these properties has become crucial in drug discovery process preventing late-stage attritions that are costly and time-consuming.
We at PozeSCAF, have developed exclusive AI-enabled ADMET models with advanced deep learning and machine learning algorithms. Our ADMET models are trained on high-quality, well-curated large datasets that combine chemical diversity by including diverse structural classes, functional groups, and chemical properties and biological diversity achieved through diverse target proteins, cell types, and organisms for capturing variations in ADMET properties across different biological systems. This helps our models generalize well and identify complex patterns and relationships that may not be apparent through traditional methods. This capability allows for a more comprehensive understanding of drug interactions and outcomes, improving the overall efficiency of drug development.
Our ADMET models analyze complex relationships between molecular structures and ADMET properties. Our AI-based ADMET models are continuously updated and refined with ever-increasing new drug data. This adaptability allows the models to improve over time, incorporating new knowledge and enhancing their predictive capabilities. The ensemble model approach combines predictions from multiple models to improve the overall accuracy and robustness of ADMET parameters that include solubility, permeability, bioavailability, Herg channel binding, microsomal stability, P-glyco protein binding, Cytochrome p450 site of metabolism, hepatotoxic, mutagenicity, carcinogenicity, organ toxicity, and Adverse Drug Reactions.
Integration of ADMET models into the drug development pipeline enhances decision-making processes, reduces the risk of late-stage failures, and ultimately contributes to more efficient and cost-effective drug development. By integrating ADMET predictions into the drug discovery process, we help researchers identify potential issues in the early stages of drug development and optimize lead compounds. This allows them to modify or discard problematic candidates before investing significant resources. The optimization can involve modifying chemical structures to improve bioavailability, reduce metabolic liabilities, or address other ADMET-related concerns and help develop drugs with improved safety profiles increasing the overall success rate of drug development programs.
A step ahead, our models help in the development of personalized medicine by considering individual variations in drug metabolism, allowing for tailored drug treatments based on a patient’s specific genetic and physiological characteristics. In a nutshell, our ADMET models offer a transformative approach to drug discovery, providing accurate predictions at a faster pace and reduced costs.
2. DrugX:
The DrugX tool innovates the discovery by integrating AI with physics-based methods. The important modules are:
- Protein modeling and druggability validation
- Rapid screening and accurate docking protocols
- Molecular dynamic simulations(MDS) and Free energy perturbations (FEP)
2.1. Protein Modeling and Druggability Validation:
Our models, generated through deep neural networks trained on protein crystallographic data, facilitate the production of energetically favorable protein conformations. These models are then clustered to produce a diverse array of stable conformers, taking into account protein flexibility to minimize false positives during screening. This aspect is particularly crucial in modeling inherently flexible proteins such as intrinsically disordered proteins (IDPs) with flexible pockets. Our platform has successfully generated drug candidates for several IDPs.
Utilizing AI and computational methods within the DrugX tool, we identify druggable pockets within proteins. This includes recognizing deeper binding pockets, flat and solvent-exposed binding sites like protein-protein interactions. Our platform excels in identifying druggable allosteric pockets for challenging proteins with toxicity or efficacy issues related to ligand binding sites. Moreover, it effectively identifies cryptic pockets within flexible proteins, thus enhancing drug discovery by transforming undruggable proteins into druggable ones.
2.2. Rapid Screening and Precise Docking Protocols:
Our integrated pipeline facilitates rapid virtual screening of the Expansive Chemical space (ECS) leveraging vast computational infrastructure equipped with GPUs and parallel computing. With our capacity to screen 1.0 billion compounds within a week, our accurate scoring functions mitigate artifacts in ultra-large database screening (ULDS) and ensure the selection of high-quality lead compounds. Additionally, our precise docking incorporates flexible binding pockets and rescoring with implicit solvent models, enhancing the enrichment of active hits.
2.3. Molecular Dynamic Simulations(MDS) and Free Energy Perturbations (FEP):
Molecular Dynamic Simulations(MDS): Molecular dynamics Simulations(MDS) play a crucial role in ranking compounds for drug discovery by providing detailed insights into molecular complex dynamics at the atomic level. These simulations capture molecular motions over time, revealing potential conformational changes upon ligand binding and solvent effects on target-ligand interactions. They generate energy profiles and accurately calculate binding affinity, aiding in the precise ranking of molecules before synthesis and testing, despite their time-consuming nature.
PozeSCAF’s GPU infrastructure and optimized parallel processing enable simulations of 200 complexes for 100ns within 2 days. In our simulation pipeline, we integrate Quantum Mechanics/Molecular Mechanics (QM/MM), utilizing quantum mechanics for active sites and classical MD for the rest of the protein. This approach accurately calculates binding affinity for charge-driven bindings like metalloproteins, and facilitates the study of enzymatic reactions and protein design. Replica Exchange Molecular Dynamics (REMD) explores protein conformational changes upon ligand binding and the effect of mutations on protein-ligand complex flexibility. Metadynamics aids in understanding free energy landscapes and kinetics of binding and unbinding (Kon and Koff), while mixed solvent sampling explores potential binding sites on target proteins, guiding rational compound design.
Free energy perturbations (FEP): Free Energy Perturbations (FEP) are instrumental in the drug discovery pipeline, particularly during hit-to-lead and lead optimization stages. These methods, such as Relative Binding Free Energy (RBFE) perturbation, aid medicinal chemists in selecting compounds for synthesis and biological activity testing. Despite their computational demands, FEP can significantly expedite lead optimization and reduce resource-intensive experimental efforts. At PozeSCAF, we use FEP simulations to extract valuable insights that inform lead optimization and enhance the potency of drug candidates. Our cutting-edge computational facility enables exhaustive and computationally intensive FEP simulations to be completed within days.
AxDrug Platform Case studies
Target Success: AxDrug has pursued 30 diverse targets, boasting a remarkable 100% success rate in unearthing potential drug candidates.
Pioneering Molecules: AxDrug has spearheaded the identification of pioneering molecules, particularly for the most challenging and intricate targets.
Innovative Polypharmacology: AxDrug has achieved significant breakthroughs by innovating best-in-class molecules with polypharmacological properties, enhancing their effectiveness across multiple targets.
Advanced Structural Modeling: Utilizing advanced techniques, AxDrug has successfully modeled intricate 3D structures and generated promising drug candidates specifically tailored for highly disordered proteins.
Proof of Concept: AxDrug has solidified the viability of each program by synthesizing approximately 200 molecules, establishing a robust proof of concept for their initiatives.
In conclusion, PozeSCAF’s AxDrug platform represents a transformative paradigm shift in drug discovery, epitomizing the synergy between AI and computational chemistry. Through its robust capabilities and comprehensive suite of tools, AxDrug empowers researchers to navigate the complexities of drug development with confidence and precision, paving the way for the next generation of life-saving medications. As technology continues to evolve, AxDrug remains at the forefront of innovation, driving advancements in pharmaceutical research and revolutionizing the way novel therapeutics are discovered, developed, and optimized.
Join us in this journey of transformation and witness the evolution of small molecule drug discovery with PozeSCAF at the helm.