Chemokine Receptors—Structure-Based Virtual Screening Assisted by Machine Learning

Abstract

Chemokines modulate the immune response by regulating the migration of immune cells. They are also known to participate in such processes as cell–cell adhesion, allograft rejection, and angiogenesis. Chemokines interact with two different subfamilies of G protein-coupled receptors: conventional chemokine receptors and atypical chemokine receptors. Here, we focused on the former one which has been linked to many inflammatory diseases, including: multiple sclerosis, asthma, nephritis, and rheumatoid arthritis. Available crystal and cryo-EM structures and homology models of six chemokine receptors (CCR1 to CCR6) were described and tested in terms of their usefulness in structure-based drug design. As a result of structure-based virtual screening for CCR2 and CCR3, several new active compounds were proposed. Known inhibitors of CCR1 to CCR6, acquired from ChEMBL, were used as training sets for two machine learning algorithms in ligand-based drug design. Performance of LightGBM was compared with a sequential Keras/TensorFlow model of neural network for these diverse datasets. A combination of structure-based virtual screening with machine learning allowed to propose several active ligands for CCR2 and CCR3 with two distinct compounds predicted as CCR3 actives by all three tested methods: Glide, Keras/TensorFlow NN, and LightGBM. In addition, the performance of these three methods in the prediction of the CCR2/CCR3 receptor subtype selectivity was assessed.

Keras/TensorFlow in Drug Design for Immunity Disorders

“…2 and 3). It means that like previously [54], predictions made by the NN model are selective for the receptor subtype because they are based on known active ligands only and not on the receptor structures which could be too similar, e.g., for SBVS. Thus, diverse CCR2, CCR3, or CXCR3 ligand training sets will provide diverse novel chemotypes for each of these receptors, while the similar structures of these receptors could only provide similar compounds in SBVS.…”

“…This means, that the CCR3 and CXCR3 NN models were less selective with respective to each other in the activity predictions for these compounds in comparison to the CCR2 model. This again suggests the dependency of NN on the training dataset composition [54], yet in this case with the desired outcome.…”

“…This means that the current datasets include only CB1 or CB2 small-molecule inhibitors and not all CB1 and CB2 actives like previousely. Furthermore, the previous datasets did not include any inactive or weakly active compounds (pChEMBL < 4), the addition of which to training sets was recently discussed in [54]. In the current datasets, nearly 40 % and 25 % (CB1, CB2, respectively) of compounds were inactive compounds (pChEMBL equal to 0).…”

“…Among ML methods, NNs or deep learning NNs built on the Keras/TensorFlow platform have been used so far mainly in binary classification tasks in drug design [91]. Here, we showed that NNs can also be used in drug design as efficient multi-class classifiers when trained on the datasets with discrete compound activity values [54]. To our knowledge, this is the first such application of Keras/TensorFlow NNs.…”

“…Furthermore, CNNs have also been used in the prediction of binding poses and affinities, and more robust models can be built by combining them with transfer and multitask learning [50]. Noteworthily, deep learning methods are not always better than those based on classical machine learning [54]. Classical ML methods are typically used for rescoring or ranking of the output from popular molecular docking programs rather than being directly integrated into them, and their results are not easily interpretable [50].…”