The rapid and accurate in silico prediction of protein-ligand binding free energies or binding affinities has the potential to transform drug discovery. In recent years, there has been a rapid growth of interest in deep learning methods for the prediction of protein-ligand binding affinities based on the structural information of protein-ligand complexes. These structure-based scoring functions often obtain better results than classical scoring functions when applied within their applicability domain. Here we review structure-based scoring functions for binding affinity prediction based on deep learning, focussing on different types of architectures, featurization strategies, data sets, methods for training and evaluation, and the role of explainable artificial intelligence in building useful models for real drug-discovery applications.

DNA-binding proteins utilise different recognition mechanisms to locate their DNA targets; some proteins recognise specific DNA sequences, while others interact with specific DNA structures. While sequence-specific DNA binding has been studied extensively, structure-specific recognition mechanisms remain unclear. Here, we study structure-specific DNA recognition by examining the structure and dynamics of DNA polymerase I Klenow Fragment (Pol) substrates both alone and in DNA–Pol complexes. Using a docking approach based on a network of 73 distances collected using single-molecule FRET, we determined a novel solution structure of the single-nucleotide-gapped DNA–Pol binary complex. The structure resembled existing crystal structures with regards to the downstream primer-template DNA substrate, and revealed a previously unobserved sharp bend (∼120°) in the DNA substrate; this pronounced bend was present in living cells. MD simulations and single-molecule assays also revealed that 4–5 nt of downstream gap-proximal DNA are unwound in the binary complex. Further, experiments and coarse-grained modelling showed the substrate alone frequently adopts bent conformations with 1–2 nt fraying around the gap, suggesting a mechanism wherein Pol recognises a pre-bent, partially-melted conformation of gapped DNA. We propose a general mechanism for substrate recognition by structure-specific enzymes driven by protein sensing of the conformational dynamics of their DNA substrates.

The Trypanosoma cruzi (T. cruzi) parasite is the cause of Chagas disease, a neglected disease endemic in South America. The life cycle of the T. cruzi parasite is complex and includes transitions between distinct life stages. This change in phenotype (without a change in genotype) could be controlled by epigenetic regulation, and might involve the bromodomain-containing factors 1−5 (TcBDF1−5). However, little is known about the function of the TcBDF1−5. Here we describe a fragment-based approach to identify ligands for T. cruzi bromodomain-containing factor 3 (TcBDF3). We expressed a soluble construct of TcBDF3 in E. coli, and used this to develop a range of biophysical assays for this protein. Fragment screening identified 12 compounds that bind to the TcBDF3 bromodomain. On the basis of this screen, we developed functional ligands containing a fluorescence or 19 F reporter group, and a photo-crosslinking probe for TcBDF3. These tool compounds will be invaluable in future studies on the function of TcBDF3 and will provide insight into the biology of T. cruzi.

sampling (REUS) molecular dynamics simulations[2]. REUS is a powerful conformational sampling method to overcome energy barriers. We applied this method to the HDAC3-T247 system and HDAC2-T247 system. These have very similar amino-acid sequences and similar structures. From results of these simulations, we obtained free energy landscapes. The results of these calculations show that the docked state is the most stable in the HDAC3-T247 system but in the HDAC2-T247 system, the docked state is less stable than undocked states. These results agree with experimental data. In this poster, we present these results. References: [1] T.

Amino acid transport into the cell is often coupled to the proton electrochemical gradient, as found in the solute carrier 36 family of proton-coupled amino acid transporters. Although no structure of a human proton-coupled amino acid transporter exists, the crystal structure of a related homolog from bacteria, GkApcT, has recently been solved in an inward-occluded state and allows an opportunity to examine how protons are coupled to amino acid transport. Our working hypothesis is that release of the amino acid substrate is facilitated by the deprotonation of a key glutamate residue (E115) located at the bottom of the binding pocket, which forms part of the intracellular gate, allowing the protein to transition from an inward-occluded to an inward-open conformation. During unbiased molecular dynamics simulations, we observed a transition from the inwardoccluded state captured in the crystal structure to a much more open state, which we consider likely to be representative of the inward-open state associated with substrate release. To explore this and the role of protons in these transitions, we have used umbrella sampling to demonstrate that the transition from inward occluded to inward open is more energetically favorable when E115 is deprotonated. That E115 is likely to be protonated in the inward-occluded state and deprotonated in the inwardopen state is further confirmed via the use of absolute binding free energies. Finally, we also show, via the use of absolute binding free energy calculations, that the affinity of the protein for alanine is very similar regardless of either the conformational state or the protonation of E115, presumably reflecting the fact that all the key interactions are deep within the binding cavity. Together, our results give a detailed picture of the role of protons in driving one of the major transitions in this transporter.