A new machine learning technique, DeepBAR, has been developed that speeds up calculations of the binding affinity of a drug molecule with a target protein.
Binding free energy
The affinity between a drug molecule and its target protein is measured by a quality known as the binding free energy. A lower binding free energy means the drug can compete better against other molecules to bind with its target. This means the drug can more effectively disrupt the proteins’ normal function. Calculating the binding free energy of a drug candidate provides an indicator of a drug’s potential efficacy. However, it is a difficult quality to assess.
Existing methods for calculating binding free energy fall into two categories.
The first category involves calculating the exact quantity, which relies on extensive molecular dynamic simulations. This approach requires simulations of both the two end states (bound and unbound) and many alchemical/physical states that bridge them. This is extremely time consuming and has significant computational cost associated with it. Approaches such as alchemical double decoupling (ADD) and potential mean force (PMF) fall into this category.
The second existing category is less computationally expensive, but yields only an approximation of the binding free energy. This method circumvents the need for intermediate states and requires sampling from the two end states only. Examples of methods that fall into this category include, molecular mechanics/Poisson-Boltzmann surface area continuum solvation (MM/PBSA), and molecular mechanics/generalized Born surface area continuum solvation (MM/GBSA).
The researchers behind this study developed a new method – DeepBAR – that combines chemistry and machine learning. Moreover, DeepBAR might help to overcome the hurdles faced by existing binding free energy calculation methods.
DeepBAR computes binding free energy exactly, but requires a fraction of the computational demands of existing methods mentioned above. The technique combines traditional chemical calculations and recent advances in machine learning (ML).
The ‘BAR’ in the name DeepBAR stands for the ‘Bennett acceptance ratio”, an algorithm that is used in exact calculations of binding free energy. Use of the Bennett acceptance ratio usually requires knowledge of the two endpoint states (the bound and unbound state), as well as knowledge of intermediate states, or states of partial binding. As mentioned previously, this process is extremely time-consuming. To overcome this, DeepBAR eliminates the need for knowledge of the intermediate states by using the Bennett acceptance ratio in the ML framework known as the deep generative model. The model creates a reference state for each endpoint. These two states are similar enough that the Bennet acceptance ratio can be used directly to identify the free binding energy, without the need for the intermediate states.
A faster future for drug screening
When tested using small protein-like molecules, DeepBAR calculated binding free energy almost 50 times faster than previous methods. The researchers hope that DeepBAR will be used in drug screening, and to aid protein design and engineering, since the method has the potential to model interactions between multiple proteins. Moving forward, the researchers hope to be able to improve DeepBAR’s ability to run calculations for larger proteins. This is being made feasible by advances in computer science.
The DeepBAR technique can quickly calculate binding affinities between drug candidates and their targets. The approach yields precise calculations in a fraction of the time compared to previous methods. It is hoped that DeepBAR may quicken the pace of drug discovery and protein engineering.
Image credit: FreePik.com