No RESP for the wicked: new AI model accelerates antibody discovery
Researchers have developed a new AI tool that predicts tight-binding antibodies and can be used to streamline antibody discovery efforts.
Antibodies’ ability to bind specifically and tightly to target biomolecules makes them ideal therapeutics; however, the process to develop these drugs can be long and expensive, and many antibodies still fail at clinical trial stages. Now, researchers from the University of California San Diego School of Medicine (CA, USA) have developed an artificial intelligence (AI)-based strategy that can identify high-affinity antibodies. The approach was used to find an antibody that binds to a cancer target 17-fold tighter than a current antibody drug. This AI pipeline, called RESP, could be used to accelerate the discovery and development of novel drugs.
Successful antibody drugs need to bind tightly to their target molecule. High-affinity antibodies are often identified through a method called directed evolution, where a known antibody sequence is mutated to create a library of antibody variants, which are then screened – using techniques such as flow cytometry – to find the tightest-binding antibodies. The selected antibodies undergo another round of mutations and screening, and this process is repeated until a set of tightly binding antibodies emerges. The repeated rounds of mutagenesis and screening make this an expensive and time-consuming process.
To streamline and accelerate the discovery process, the researchers developed the RESP AI model. The approach starts similarly, with an initial round of mutagenesis and screening; however, instead of repeating the process over and over again, the researchers feed the resulting data into a Bayesian neural network. The model analyzes the data and uses it to predict the binding affinity of other sequences.
Functional MRI reveals how time in space changes our brains
Our brains are optimized to live with gravity, but how does it adapt to extended periods of time in space, experiencing different laws of physics?
“With our machine learning tools, these subsequent rounds of sequence mutation and selection can be carried out quickly and efficiently on a computer rather than in the lab,” explained senior author Wei Wang.
As a proof of concept, the group used the model to design an antibody against programmed death ligand 1 (PD-L1), a protein that plays a key role in cancer and is the target of several anti-cancer drugs. They found an antibody that bound PD-L1 17-fold tighter than atezolizumab, an antibody drug approved for the treatment of a number of cancers.
The researchers are now developing AI models that assess other important antibody properties, such as stability, solubility and selectivity. “By combining these AI tools, scientists may be able to perform an increasing share of their antibody discovery efforts on a computer instead of at the bench, potentially leading to a faster and less failure-prone discovery process,” said Wang. “There are so many applications to this pipeline, and these findings are really just the beginning.”
