Crystal Structure Prediction
I: 2D representation
II: Stable conformers
III: Stable polymorphs
Context
Accelerated workflow for a free energy landscape scan
Such an undertaking requires implementing computationally expensive Quantum mechanical/chemical methods. This expense is compounded multifold if the polymorph studies include salts, cocrystals, hydrates/solvates, etc.
Aganitha utilizes advanced AI/ML models in tandem with GPU-based QM software, all running on elastic/scalable cloud infrastructure, to aid in exploring the energetics of a range of candidate crystal structures within shorter time frames. These can serve as valuable computational aids in the formulation efforts of your product.
Our Solution
Scalable and accelerated in silico pipeline to identify stable polymorphs
- Generative models to navigate the conformational space of an organic molecule
- Diffusion based models to generate candidate crystal structures
- Graph Neural Networks (GNN) based models to predict Lattice Energy
- Accelerated DFT computational pipelines to screen candidate crystal structures
Key components of Aganitha's CSP Pipeline
Accurately predicting the different polymorphs of an Active Pharmaceutical Ingredient (API), is a key part of drug development. Conformer ensemble, which is a set of stable 3D conformations of a molecule are key inputs to CSP. Lattice Energy (LE) predictions using AI/ML-based models and accelerated DFT-based calculations are useful in the preliminary screening of CSP workflows.
Solution area
Our Innovative Solution: MolConSUL
Introducing MolConSUL (Molecular Conformer Search with Unsupervised Learning)—Aganitha’s proprietary tool designed to overcome the limitations of traditional methods. MolConSUL optimizes conformer generation, balancing accuracy with speed to streamline your CSP workflows.
Key Features:
- Computationally affordable: Thoroughly benchmarked to achieve the right balance between accuracy and computation time.
- Captures conformational diversity: Ensures to capture the conformational diversity of APIs with a limited number of conformers in the final ensemble.
- Customizable modular platform: Provides a flexible platform that can be tailored to meet specific user requirements.
Solution area
Optimizing Lattice Energy Calculations: Importance and Challenges
In the context of CSP, the challenge is to accurately rank order a million of possible crystal structures. Fast and accurate LE prediction is vital for efficiently ranking and identifying the low energy crystal structures. We recognize the need and potential of accurate Deep Learning (DL) models in predicting LE for organic molecular crystals. .