De novo Drug Design

Medicinal chemists are more often required to change molecular frameworks or scaffolds further by so-called core hopping to address scaffold-dependent issues. Artificial intelligence(AI) equips medicinal chemistry with innovative tools for small molecular design and lead discovery. AI-driven de novo drug design aims to generate new chemical entities with desired properties in a cost- and time-efficient manner. The ability of approaches provided by CD ComputaBio's AI platform to generate innovative molecular cores has been proved, thereby exploring novel regions of the chemical space.

Generative AI Model

The generative AI model has successfully produced molecules that possess features of the synthetic ChEMBL compounds used for model training. The model not only produced a high proportion of valid, stable and innovative structures but also captured the bioactivity of the templates.

Rely on generative artificial intelligence technologies.
Learn from known bioactive compounds database.
Autonomously design novel compounds with inherited bioactivity and synthesizability.
Generate large numbers of diverse structures.

Figure 1 Generative AI Model

The AI approach consisted of several basic steps.

Develop a generic model (utilizing a recently published deep recurrent neural network) that learns the constitution of drug-like molecules from a large unfocussed compound set.
Trained a recurrent neural network to capture the constitution of a large set of known bioactive compounds represented as SMILES strings.
Fine‐tune this generic model on more specific molecular features from a small target‐focused library of actives, and then transfer learning to enable the de novo drug design and generation.
Produce novel chemical entities within the training data domain from the resulting fine‐tuned AI model.

Algorithm

Evolutionary algorithms are actively used for de novo drug design, which are based on those concepts derived from biological evolution, including reproduction, mutation (fragment-based mutation and atom-based mutation), crossover, and selection.

Computerized compound design
Parameter optimization of QSAR/QSPR models
3D-ligand alignment
Select the surviving structures using fitness score

Building structures with chemical feasibility is an important point for de novo drug design. Generate chemical structures using fragment-based approaches and utilizing the structures of known ligands obey valence rules, RECAP-based rule (11 reaction schemes) or other connection rules.

Figure 2 De novo Drug Design

Workflow

Data Input (e.g. a reference structure, an active molecule)
Generate seed fragments and initial structures.
Fragment-based mutation methods and crossover.
Evaluate fitness of the structures (Tanimoto coefficient).
Select some of them to survive (tournament method).
Iteration
Output designed structures

Available Design Programs

Developed computerized structural design approaches utilize protein-structures and/or ligand-structures as the structure-base design and ligand-based design, respectively. Site point connection method includes LUDI. Fragment connection methods include SPLICE, NEW LEAD, and PRO-LIGAND. Sequential build up methods include LEGEND, GROW, and SPROUT. Random connection and disconnection methods include CONCEPTS, CONCERTS, MCDNLG.

Figure 3 Available Programs for De novo Drug Design

Results and Evaluation

nomain-title-log-pic2 Use target prediction method (SPiDER) and molecular shape and partial charge descriptors to determine the similarity of the designed compounds to known bioactive chemicals, taking into account their individual in silico ranks and building block availability.

Benchmark fingerprint descriptors for virtual screening to determine the structural similarity.

AtomPairs fingerprints
Morgan fingerprints
RDKit fingerprints
MACCS keys

nomain-title-log-pic2 Confirm the designed compound that is not present in ChEMBL, PubChem, SureChEMBL, Reaxys and SciFinder databases.