Abstract:Protein kinases are structurally dynamic proteins that control downstream signaling cascades by phosphorylating their substrates. Protein kinases regulate their function by adopting several conformational states in their active site determined by the movements of several motifs such as the αC-Helix, DFG residues and the activation loop. Each conformational state represents a distinct physicochemical environment that accepts or precludes ligand binding. However, most of the kinome have not been crystalized across these possible conformational states. It has been shown that shallow Multiple Sequence Alignments (MSA) can enable AlphaFold2 (AF2) to model kinases in alternative conformations. However, it is unclear if these models can be leveraged for structure-based drug discovery. Additionally, there are several machine learning tools to predict protein-ligand interactions based on ligand chemotype and binding pocket properties, but these models cannot be used to identify ligands with clear state specificity. Here, we first present an approach called AlphaFold2 Steering (AF2-Steering), a systematic methodology to direct AF2 to sample kinases in the active and inactive conformations. We use our approach to model the protein kinome in precise conformational states. We demonstrate the utility of these AF2-steered kinase models by employing them in a prospective virtual screening study that integrates machine learning with docking to find state specific inhibitors for well-studied and dark kinases that lack structures in the active conformational state. We then experimentally validate the hits, an essential step often overlooked, and later experimentally confirm the conformation-specificity of the ligands identified for FLT3, a protein kinase that currently lacks an active state crystal structure. Against a strict binding criterion of at least 1μM Kd, our modelled structures achieved an overall hit rate of 53%. We also confirm the conformation-specificity of 4/7 FLT3 ligands, thus demonstrating the value of MSA-steered AF2 modelled kinase structures combined with machine learning and docking to guide conformation-specific kinase drug discovery.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to use machine learning and AlphaFold2 to generate protein kinase inhibitors in specific conformational states, especially in the absence of active - conformation crystal structures. Specifically, the researchers developed a method named AlphaFold2 Steering (AF2 - Steering), which systematically adjusts the multiple sequence alignment (MSA) to guide AlphaFold2 to generate protein kinase models in specific conformational states. Then, they use these models for virtual screening, combined with machine - learning and molecular docking techniques, to discover inhibitors that are selective for specific conformations. In addition, the study also experimentally verified the effectiveness of these models, especially for protein kinases such as FLT3 that lack crystal structures in the active conformation. ### Main Objectives: 1. **Develop the AF2 - Steering method**: Guide AlphaFold2 to generate protein kinase models in specific conformational states by adjusting MSA parameters. 2. **Generate high - coverage CIDI conformational models**: Use the AF2 - Steering method to generate CIDI conformational models that cover the entire protein kinase group (kinome). 3. **Virtual screening and experimental verification**: Use the generated models for virtual screening, combined with machine - learning and molecular docking techniques, to identify selective inhibitors for specific conformations, and experimentally verify the effectiveness and conformational specificity of these inhibitors. ### Specific Problems Solved: - **Lack of crystal structures in the active conformation**: Many protein kinases do not have crystal structures in the active conformation, which limits the application of structure - based drug design (SBDD). - **Improve the hit rate of virtual screening**: Improve the hit rate of virtual screening by combining machine - learning and molecular docking techniques, especially to discover new inhibitors in "dark kinases". - **Verify conformational specificity**: Experimentally verify whether the discovered inhibitors indeed have conformational specificity, especially in protein kinases that lack active - conformation crystal structures. ### Experimental Verification: - **Virtual screening**: Use the generated AF2 - Steering models for virtual screening, combined with machine - learning and molecular docking techniques, to identify potential inhibitors. - **Experimental verification**: Verify the binding ability and conformational specificity of these inhibitors through commercial competitive binding assays (such as KINOMEscan and scanMODE). Through these methods, the researchers not only successfully generated high - quality CIDI conformational models, but also experimentally verified the effectiveness of these models in the experiment, providing new tools and methods for drug discovery of protein kinases.

Leveraging Machine Learning and AlphaFold2 Steering to Discover State-Specific Inhibitors Across the Kinome

A comprehensive exploration of the druggable conformational space of protein kinases using AI-predicted structures

Improving docking and virtual screening performance using AlphaFold2 multi-state modeling for kinases

Applying multi-state modeling using AlphaFold2 for kinases and its application for ensemble screening

Empowering AlphaFold2 for protein conformation selective drug discovery with AlphaFold2-RAVE

Interpretable Atomistic Prediction and Functional Analysis of Conformational Ensembles and Allosteric States in Protein Kinases Using AlphaFold2 Adaptation with Randomized Sequence Scanning and Local Frustration Profiling

Docking-informed machine learning for kinome wide affinity prediction

A Hybrid Structure-Based Machine Learning Approach for Predicting Kinase Inhibition by Small Molecules

MinKLIFSAI: a simple machine learning approach toward selective kinase inhibitor

Lessons learned during the journey of data: from experiment to model for predicting kinase affinity, selectivity, polypharmacology, and resistance

Prediction of Conformational Ensembles and Structural Effects of State-Switching Allosteric Mutants in the Protein Kinases Using Comparative Analysis of AlphaFold2 Adaptations with Sequence Masking and Shallow Subsampling

Are Deep Learning Structural Models Sufficiently Accurate for Virtual Screening? Application of Docking Algorithms to AlphaFold2 Predicted Structures

Active site sequence representation of human kinases outperforms full sequence for affinity prediction and inhibitor generation: 3D effects in a 1D model

Exploring Kinase Asp-Phe-Gly (DFG) Loop Conformational Stability with AlphaFold2-RAVE

Leveraging multiple data types for improved compound-kinase bioactivity prediction

Exploring kinase DFG loop conformational stability with AlphaFold2-RAVE

Integrating Molecular Dynamics and Machine Learning Algorithms to Predict the Functional Profile of Kinase Ligands

Leveraging nonstructural data to predict structures and affinities of protein-ligand complexes

Using the Structural Kinome to Systematize Kinase Drug Discovery

Classifying protein kinase conformations with machine learning