Targeting CDK9 Reactivates Epigenetically Silenced Genes in Cancer.

Cyclin-dependent kinase 9 (CDK9) promotes transcriptional elongation through RNAPII pause release. We now report that CDK9 is also essential for maintaining gene silencing at heterochromatic loci. Through a live cell drug screen with genetic confirmation, we discovered that CDK9 inhibition reactivates epigenetically silenced genes in cancer, leading to restored tumor suppressor gene expression, cell differentiation, and activation of endogenous retrovirus genes. CDK9 inhibition dephosphorylates the SWI/SNF protein BRG1, which contributes to gene reactivation. By optimization through gene expression, we developed a highly selective CDK9 inhibitor (MC180295, IC50 = 5 nM) that has broad anti-cancer activity in vitro and is effective in in vivo cancer models. Additionally, CDK9 inhibition sensitizes to the immune checkpoint inhibitor α-PD-1 in vivo, making it an excellent target for epigenetic therapy of cancer.

Macromolecular modeling and design in Rosetta: recent methods and frameworks.

The Rosetta software for macromolecular modeling, docking and design is extensively used in laboratories worldwide. During two decades of development by a community of laboratories at more than 60 institutions, Rosetta has been continuously refactored and extended. Its advantages are its performance and interoperability between broad modeling capabilities. Here we review tools developed in the last 5 years, including over 80 methods. We discuss improvements to the score function, user interfaces and usability. Rosetta is available at http://www.rosettacommons.org.

The role of NSD1, NSD2, and NSD3 histone methyltransferases in solid tumors.

NSD1, NSD2, and NSD3 constitute the nuclear receptor-binding SET Domain (NSD) family of histone 3 lysine 36 (H3K36) methyltransferases. These structurally similar enzymes mono- and di-methylate H3K36, which contribute to the maintenance of chromatin integrity and regulate the expression of genes that control cell division, apoptosis, DNA repair, and epithelial-mesenchymal transition (EMT). Aberrant expression or mutation of members of the NSD family is associated with developmental defects and the occurrence of some types of cancer. In this review, we discuss the effect of alterations in NSDs on cancer patient's prognosis and response to treatment. We summarize the current understanding of the biological functions of NSD proteins, focusing on their activities and the role in the formation and progression in solid tumors biology, as well as how it depends on tumor etiologies. This review also discusses ongoing efforts to develop NSD inhibitors as a promising new class of cancer therapeutic agents.

Machine learning classification can reduce false positives in structure-based virtual screening.

With the recent explosion in the size of libraries available for screening, virtual screening is positioned to assume a more prominent role in early drug discovery's search for active chemical matter. In typical virtual screens, however, only about 12% of the top-scoring compounds actually show activity when tested in biochemical assays. We argue that most scoring functions used for this task have been developed with insufficient thoughtfulness into the datasets on which they are trained and tested, leading to overly simplistic models and/or overtraining. These problems are compounded in the literature because studies reporting new scoring methods have not validated their models prospectively within the same study. Here, we report a strategy for building a training dataset (D-COID) that aims to generate highly compelling decoy complexes that are individually matched to available active complexes. Using this dataset, we train a general-purpose classifier for virtual screening (vScreenML) that is built on the XGBoost framework. In retrospective benchmarks, our classifier shows outstanding performance relative to other scoring functions. In a prospective context, nearly all candidate inhibitors from a screen against acetylcholinesterase show detectable activity; beyond this, 10 of 23 compounds have IC50 better than 50 µM. Without any medicinal chemistry optimization, the most potent hit has IC50 280 nM, corresponding to Ki of 173 nM. These results support using the D-COID strategy for training classifiers in other computational biology tasks, and for vScreenML in virtual screening campaigns against other protein targets. Both D-COID and vScreenML are freely distributed to facilitate such efforts.

A clinically relevant polymorphism in the Na+/taurocholate cotransporting polypeptide (NTCP) occurs at a rheostat position.

Conventionally, most amino acid substitutions at "important" protein positions are expected to abolish function. However, in several soluble-globular proteins, we identified a class of nonconserved positions for which various substitutions produced progressive functional changes; we consider these evolutionary "rheostats". Here, we report a strong rheostat position in the integral membrane protein, Na+/taurocholate (TCA) cotransporting polypeptide, at the site of a pharmacologically relevant polymorphism (S267F). Functional studies were performed for all 20 substitutions (S267X) with three substrates (TCA, estrone-3-sulfate, and rosuvastatin). The S267X set showed strong rheostatic effects on overall transport, and individual substitutions showed varied effects on transport kinetics (Km and Vmax) and substrate specificity. To assess protein stability, we measured surface expression and used the Rosetta software (https://www.rosettacommons.org) suite to model structure and stability changes of S267X. Although buried near the substrate-binding site, S267X substitutions were easily accommodated in the Na+/TCA cotransporting polypeptide structure model. Across the modest range of changes, calculated stabilities correlated with surface-expression differences, but neither parameter correlated with altered transport. Thus, substitutions at rheostat position 267 had wide-ranging effects on the phenotype of this integral membrane protein. We further propose that polymorphic positions in other proteins might be locations of rheostat positions.

Structural Plasticity Is a Feature of Rheostat Positions in the Human Na+/Taurocholate Cotransporting Polypeptide (NTCP).

In the Na+/taurocholate cotransporting polypeptide (NTCP), the clinically relevant S267F polymorphism occurs at a "rheostat position". That is, amino acid substitutions at this position ("S267X") lead to a wide range of functional outcomes. This result was particularly striking because molecular models predicted the S267X side chains are buried, and thus, usually expected to be less tolerant of substitutions. To assess whether structural tolerance to buried substitutions is widespread in NTCP, here we used Rosetta to model all 19 potential substitutions at another 13 buried positions. Again, only subtle changes in the calculated stabilities and structures were predicted. Calculations were experimentally validated for 19 variants at codon 271 ("N271X"). Results showed near wildtype expression and rheostatic modulation of substrate transport, implicating N271 as a rheostat position. Notably, each N271X substitution showed a similar effect on the transport of three different substrates and thus did not alter substrate specificity. This differs from S267X, which altered both transport kinetics and specificity. As both transport and specificity may change during protein evolution, the recognition of such rheostat positions may be important for evolutionary studies. We further propose that the presence of rheostat positions is facilitated by local plasticity within the protein structure. Finally, we note that identifying rheostat positions may advance efforts to predict new biomedically relevant missense variants in NTCP and other membrane transport proteins.

Efficient Hit-to-Lead Searching of Kinase Inhibitor Chemical Space via Computational Fragment Merging.

In early-stage drug discovery, the hit-to-lead optimization (or "hit expansion") stage entails starting from a newly identified active compound and improving its potency or other properties. Traditionally, this process relies on synthesizing and evaluating a series of analogues to build up structure-activity relationships. Here, we describe a computational strategy focused on kinase inhibitors, intended to expedite the process of identifying analogues with improved potency. Our protocol begins from an inhibitor of the target kinase and generalizes the synthetic route used to access it. By searching for commercially available replacements for the individual building blocks used to make the parent inhibitor, we compile an enumerated library of compounds that can be accessed using the same chemical transformations; these huge libraries can exceed many millions─or billions─of compounds. Because the resulting libraries are much too large for explicit virtual screening, we instead consider alternate approaches to identify the top-scoring compounds. We find that contributions from individual substituents are well described by a pairwise additivity approximation, provided that the corresponding fragments position their shared core in precisely the same way relative to the binding site. This key insight allows us to determine which fragments are suitable for merging into single new compounds and which are not. Further, the use of pairwise approximation allows interaction energies to be assigned to each compound in the library without the need for any further structure-based modeling: interaction energies instead can be reliably estimated from the energies of the component fragments, and the reduced computational requirements allow for flexible energy minimizations that allow the kinase to respond to each substitution. We demonstrate this protocol using libraries built from six representative kinase inhibitors drawn from the literature, which target five different kinases: CDK9, CHK1, CDK2, EGFRT790M, and ACK1. In each example, the enumerated library includes additional analogues reported by the original study to have activity, and these analogues are successfully prioritized within the library. We envision that the insights from this work can facilitate the rapid assembly and screening of increasingly large libraries for focused hit-to-lead optimization. To enable adoption of these methods and to encourage further analyses, we disseminate the computational tools needed to deploy this protocol.

Rationalizing PROTAC-Mediated Ternary Complex Formation Using Rosetta.

Proteolysis-targeting chimaeras (PROTACs) are molecules that combine a target-binding warhead with an E3 ligase-recruiting moiety; by drawing the target protein into a ternary complex with the E3 ligase, PROTACs induce target protein degradation. While PROTACs hold exciting potential as chemical probes and as therapeutic agents, development of a PROTAC typically requires synthesis of numerous analogs to thoroughly explore variations on the chemical linker; without extensive trial and error, it is unclear how to link the two protein-recruiting moieties to promote formation of a productive ternary complex. Here, we describe a structure-based computational method for evaluating the suitability of a given linker for ternary complex formation. Our method uses Rosetta to dock the protein components and then builds the PROTAC from its component fragments into each binding mode; complete models of the ternary complex are then refined. We apply this approach to retrospectively evaluate multiple PROTACs from the literature, spanning diverse target proteins. We find that modeling ternary complex formation is sufficient to explain both activity and selectivity reported for these PROTACs, implying that other cellular factors are not key determinants of activity in these cases. We further find that interpreting PROTAC activity is best approached using an ensemble of structures of the ternary complex rather than a single static conformation and that members of a structurally conserved protein family can be recruited by the same PROTAC through vastly different binding modes. To encourage adoption of these methods and promote further analyses, we disseminate both the computational methods and the models of ternary complexes.

A de novo protein binding pair by computational design and directed evolution.

The de novo design of protein-protein interfaces is a stringent test of our understanding of the principles underlying protein-protein interactions and would enable unique approaches to biological and medical challenges. Here we describe a motif-based method to computationally design protein-protein complexes with native-like interface composition and interaction density. Using this method we designed a pair of proteins, Prb and Pdar, that heterodimerize with a Kd of 130 nM, 1000-fold tighter than any previously designed de novo protein-protein complex. Directed evolution identified two point mutations that improve affinity to 180 pM. Crystal structures of an affinity-matured complex reveal binding is entirely through the designed interface residues. Surprisingly, in the in vitro evolved complex one of the partners is rotated 180° relative to the original design model, yet still maintains the central computationally designed hotspot interaction and preserves the character of many peripheral interactions. This work demonstrates that high-affinity protein interfaces can be created by designing complementary interaction surfaces on two noninteracting partners and underscores remaining challenges.

Promysalin Elicits Species-Selective Inhibition of Pseudomonas aeruginosa by Targeting Succinate Dehydrogenase.

Natural products have served as an inspiration to scientists both for their complex three-dimensional architecture and exquisite biological activity. Promysalin is one such Pseudomonad secondary metabolite that exhibits narrow-spectrum antibacterial activity, originally isolated from the rhizosphere. We herein utilize affinity-based protein profiling (AfBPP) to identify succinate dehydrogenase (Sdh) as the biological target of the natural product. The target was further validated in silico, in vitro, in vivo, and through the selection, and sequencing, of a resistant mutant. Succinate dehydrogenase plays an essential role in primary metabolism of Pseudomonas aeruginosa as the only enzyme that is involved both in the tricarboxylic acid cycle (TCA) and in respiration via the electron transport chain. These findings add credence to other studies that suggest that the TCA cycle is an understudied target in the development of novel therapeutics to combat P. aeruginosa, a significant pathogen in clinical settings.

Natural product derivative Gossypolone inhibits Musashi family of RNA-binding proteins.

BACKGROUND: The Musashi (MSI) family of RNA-binding proteins is best known for the role in post-transcriptional regulation of target mRNAs. Elevated MSI1 levels in a variety of human cancer are associated with up-regulation of Notch/Wnt signaling. MSI1 binds to and negatively regulates translation of Numb and APC (adenomatous polyposis coli), negative regulators of Notch and Wnt signaling respectively. METHODS: Previously, we have shown that the natural product (-)-gossypol as the first known small molecule inhibitor of MSI1 that down-regulates Notch/Wnt signaling and inhibits tumor xenograft growth in vivo. Using a fluorescence polarization (FP) competition assay, we identified gossypolone (Gn) with a > 20-fold increase in Ki value compared to (-)-gossypol. We validated Gn binding to MSI1 using surface plasmon resonance, nuclear magnetic resonance, and cellular thermal shift assay, and tested the effects of Gn on colon cancer cells and colon cancer DLD-1 xenografts in nude mice. RESULTS: In colon cancer cells, Gn reduced Notch/Wnt signaling and induced apoptosis. Compared to (-)-gossypol, the same concentration of Gn is less active in all the cell assays tested. To increase Gn bioavailability, we used PEGylated liposomes in our in vivo studies. Gn-lip via tail vein injection inhibited the growth of human colon cancer DLD-1 xenografts in nude mice, as compared to the untreated control (P < 0.01, n = 10). CONCLUSION: Our data suggest that PEGylation improved the bioavailability of Gn as well as achieved tumor-targeted delivery and controlled release of Gn, which enhanced its overall biocompatibility and drug efficacy in vivo. This provides proof of concept for the development of Gn-lip as a molecular therapy for colon cancer with MSI1/MSI2 overexpression.

Using homology modeling to interrogate binding affinity in neutralization of ricin toxin by a family of single domain antibodies.

In this report we investigated, within a group of closely related single domain camelid antibodies (VH Hs), the relationship between binding affinity and neutralizing activity as it pertains to ricin, a fast-acting toxin and biothreat agent. The V1C7-like VH Hs (V1C7, V2B9, V2E8, and V5C1) are similar in amino acid sequence, but differ in their binding affinities and toxin-neutralizing activities. Using the X-ray crystal structure of V1C7 in complex with ricin's enzymatic subunit (RTA) as a template, Rosetta-based homology modeling coupled with energetic decomposition led us to predict that a single pairwise interaction between Arg29 on V5C1 and Glu67 on RTA was responsible for the difference in ricin toxin binding affinity between V1C7, a weak neutralizer, and V5C1, a moderate neutralizer. This prediction was borne out experimentally: substitution of Arg for Gly at position 29 enhanced V1C7's binding affinity for ricin, whereas the reverse (ie, Gly for Arg at position 29) diminished V5C1's binding affinity by >10 fold. As expected, the V5C1R29G mutant was largely devoid of toxin-neutralizing activity (TNA). However, the TNA of the V1C7G29R mutant was not correspondingly improved, indicating that in the V1C7 family binding affinity alone does not account for differences in antibody function. V1C7 and V5C1, as well as their respective point mutants, recognized indistinguishable epitopes on RTA, at least at the level of sensitivity afforded by hydrogen-deuterium mass spectrometry. The results of this study have implications for engineering therapeutic antibodies because they demonstrate that even subtle differences in epitope specificity can account for important differences in antibody function.

"Solvent hydrogen-bond occlusion": A new model of polar desolvation for biomolecular energetics.

Water engages in two important types of interactions near biomolecules: it forms ordered "cages" around exposed hydrophobic regions, and it participates in hydrogen bonds with surface polar groups. Both types of interaction are critical to biomolecular structure and function, but explicitly including an appropriate number of solvent molecules makes many applications computationally intractable. A number of implicit solvent models have been developed to address this problem, many of which treat these two solvation effects separately. Here, we describe a new model to capture polar solvation effects, called SHO ("solvent hydrogen-bond occlusion"); our model aims to directly evaluate the energetic penalty associated with displacing discrete first-shell water molecules near each solute polar group. We have incorporated SHO into the Rosetta energy function, and find that scoring protein structures with SHO provides superior performance in loop modeling, virtual screening, and protein structure prediction benchmarks. These improvements stem from the fact that SHO accurately identifies and penalizes polar groups that do not participate in hydrogen bonds, either with solvent or with other solute atoms ("unsatisfied" polar groups). We expect that in future, SHO will enable higher-resolution predictions for a variety of molecular modeling applications. © 2017 Wiley Periodicals, Inc.

Optimal allosteric stabilization sites using contact stabilization analysis.

Proteins can be destabilized by a number of environmental factors such as temperature, pH, and mutation. The ability to subsequently restore function under these conditions by adding small molecule stabilizers, or by introducing disulfide bonds, would be a very powerful tool, but the physical principles that drive this stabilization are not well understood. The first problem lies is in choosing an appropriate binding site or disulfide bond location to best confer stability to the active site and restore function. Here, we present a general framework for predicting which allosteric binding sites correlate with stability in the active site. Using the Karanicolas-Brooks Go-like model, we examine the dynamics of the enzyme ß-glucuronidase using an Umbrella Sampling method to thoroughly sample the conformational landscape. Each intramolecular contact is assigned a score termed a "stabilization factor" that measures its correlation with structural changes in the active site. We have carried out this analysis for three different scaling strengths for the intramolecular contacts, and we examine how the calculated stabilization factors depend on the ensemble of destabilized conformations. We further examine a locally destabilized mutant of ß-glucuronidase that has been characterized experimentally, and show that this brings about local changes in the stabilization factors. We find that the proximity to the active site is not sufficient to determine which contacts can confer active site stability. © 2016 Wiley Periodicals, Inc.

Structure-based design of non-natural amino-acid inhibitors of amyloid fibril formation.

Many globular and natively disordered proteins can convert into amyloid fibrils. These fibrils are associated with numerous pathologies as well as with normal cellular functions, and frequently form during protein denaturation. Inhibitors of pathological amyloid fibril formation could be useful in the development of therapeutics, provided that the inhibitors were specific enough to avoid interfering with normal processes. Here we show that computer-aided, structure-based design can yield highly specific peptide inhibitors of amyloid formation. Using known atomic structures of segments of amyloid fibrils as templates, we have designed and characterized an all-D-amino-acid inhibitor of the fibril formation of the tau protein associated with Alzheimer's disease, and a non-natural L-amino-acid inhibitor of an amyloid fibril that enhances sexual transmission of human immunodeficiency virus. Our results indicate that peptides from structure-based designs can disrupt the fibril formation of full-length proteins, including those, such as tau protein, that lack fully ordered native structures. Because the inhibiting peptides have been designed on structures of dual-ß-sheet 'steric zippers', the successful inhibition of amyloid fibril formation strengthens the hypothesis that amyloid spines contain steric zippers.

Chaperonin-Based Biolayer Interferometry To Assess the Kinetic Stability of Metastable, Aggregation-Prone Proteins.

Stabilizing the folded state of metastable and/or aggregation-prone proteins through exogenous ligand binding is an appealing strategy for decreasing disease pathologies caused by protein folding defects or deleterious kinetic transitions. Current methods of examining binding of a ligand to these marginally stable native states are limited because protein aggregation typically interferes with analysis. Here, we describe a rapid method for assessing the kinetic stability of folded proteins and monitoring the effects of ligand stabilization for both intrinsically stable proteins (monomers, oligomers, and multidomain proteins) and metastable proteins (e.g., low Tm) that uses a new GroEL chaperonin-based biolayer interferometry (BLI) denaturant pulse platform. A kinetically controlled denaturation isotherm is generated by exposing a target protein, immobilized on a BLI biosensor, to increasing denaturant concentrations (urea or GuHCl) in a pulsatile manner to induce partial or complete unfolding of the attached protein population. Following the rapid removal of the denaturant, the extent of hydrophobic unfolded/partially folded species that remains is detected by an increased level of GroEL binding. Because this kinetic denaturant pulse is brief, the amplitude of binding of GroEL to the immobilized protein depends on the duration of the exposure to the denaturant, the concentration of the denaturant, wash times, and the underlying protein unfolding-refolding kinetics; fixing all other parameters and plotting the GroEL binding amplitude versus denaturant pulse concentration result in a kinetically controlled denaturation isotherm. When folding osmolytes or stabilizing ligands are added to the immobilized target proteins before and during the denaturant pulse, the diminished population of unfolded/partially folded protein manifests as a decreased level of GroEL binding and/or a marked shift in these kinetically controlled denaturation profiles to higher denaturant concentrations. This particular platform approach can be used to identify small molecules and/or solution conditions that can stabilize or destabilize thermally stable proteins, multidomain proteins, oligomeric proteins, and, most importantly, aggregation-prone metastable proteins.

Selectivity by small-molecule inhibitors of protein interactions can be driven by protein surface fluctuations.

Small-molecules that inhibit interactions between specific pairs of proteins have long represented a promising avenue for therapeutic intervention in a variety of settings. Structural studies have shown that in many cases, the inhibitor-bound protein adopts a conformation that is distinct from its unbound and its protein-bound conformations. This plasticity of the protein surface presents a major challenge in predicting which members of a protein family will be inhibited by a given ligand. Here, we use biased simulations of Bcl-2-family proteins to generate ensembles of low-energy conformations that contain surface pockets suitable for small molecule binding. We find that the resulting conformational ensembles include surface pockets that mimic those observed in inhibitor-bound crystal structures. Next, we find that the ensembles generated using different members of this protein family are overlapping but distinct, and that the activity of a given compound against a particular family member (ligand selectivity) can be predicted from whether the corresponding ensemble samples a complementary surface pocket. Finally, we find that each ensemble includes certain surface pockets that are not shared by any other family member: while no inhibitors have yet been identified to take advantage of these pockets, we expect that chemical scaffolds complementing these "distinct" pockets will prove highly selective for their targets. The opportunity to achieve target selectivity within a protein family by exploiting differences in surface fluctuations represents a new paradigm that may facilitate design of family-selective small-molecule inhibitors of protein-protein interactions.

Ultra-High-Throughput Structure-Based Virtual Screening for Small-Molecule Inhibitors of Protein-Protein Interactions.

Protein-protein interactions play important roles in virtually all cellular processes, making them enticing targets for modulation by small-molecule therapeutics: specific examples have been well validated in diseases ranging from cancer and autoimmune disorders, to bacterial and viral infections. Despite several notable successes, however, overall these remain a very challenging target class. Protein interaction sites are especially challenging for computational approaches, because the target protein surface often undergoes a conformational change to enable ligand binding: this confounds traditional approaches for virtual screening. Through previous studies, we demonstrated that biased "pocket optimization" simulations could be used to build collections of low-energy pocket-containing conformations, starting from an unbound protein structure. Here, we demonstrate that these pockets can further be used to identify ligands that complement the protein surface. To do so, we first build from a given pocket its "exemplar": a perfect, but nonphysical, pseudoligand that would optimally match the shape and chemical features of the pocket. In our previous studies, we used these exemplars to quantitatively compare protein surface pockets to one another. Here, we now introduce this exemplar as a template for pharmacophore-based screening of chemical libraries. Through a series of benchmark experiments, we demonstrate that this approach exhibits comparable performance as traditional docking methods for identifying known inhibitors acting at protein interaction sites. However, because this approach is predicated on ligand/exemplar overlays, and thus does not require explicit calculation of protein-ligand interactions, exemplar screening provides a tremendous speed advantage over docking: 6 million compounds can be screened in about 15 min on a single 16-core, dual-GPU computer. The extreme speed at which large compound libraries can be traversed easily enables screening against a "pocket-optimized" ensemble of protein conformations, which in turn facilitates identification of more diverse classes of active compounds for a given protein target.

Druggable protein interaction sites are more predisposed to surface pocket formation than the rest of the protein surface.

Despite intense interest and considerable effort via high-throughput screening, there are few examples of small molecules that directly inhibit protein-protein interactions. This suggests that many protein interaction surfaces may not be intrinsically "druggable" by small molecules, and elevates in importance the few successful examples as model systems for improving our fundamental understanding of druggability. Here we describe an approach for exploring protein fluctuations enriched in conformations containing surface pockets suitable for small molecule binding. Starting from a set of seven unbound protein structures, we find that the presence of low-energy pocket-containing conformations is indeed a signature of druggable protein interaction sites and that analogous surface pockets are not formed elsewhere on the protein. We further find that ensembles of conformations generated with this biased approach structurally resemble known inhibitor-bound structures more closely than equivalent ensembles of unbiased conformations. Collectively these results suggest that "druggability" is a property encoded on a protein surface through its propensity to form pockets, and inspire a model in which the crude features of the predisposed pocket(s) restrict the range of complementary ligands; additional smaller conformational changes then respond to details of a particular ligand. We anticipate that the insights described here will prove useful in selecting protein targets for therapeutic intervention.

PPIscreenML: Structure-based screening for protein-protein interactions using AlphaFold.

Protein-protein interactions underlie nearly all cellular processes. With the advent of protein structure prediction methods such as AlphaFold2 (AF2), models of specific protein pairs can be built extremely accurately in most cases. However, determining the relevance of a given protein pair remains an open question. It is presently unclear how to use best structure-based tools to infer whether a pair of candidate proteins indeed interact with one another: ideally, one might even use such information to screen amongst candidate pairings to build up protein interaction networks. Whereas methods for evaluating quality of modeled protein complexes have been co-opted for determining which pairings interact (e.g., pDockQ and iPTM), there have been no rigorously benchmarked methods for this task. Here we introduce PPIscreenML, a classification model trained to distinguish AF2 models of interacting protein pairs from AF2 models of compelling decoy pairings. We find that PPIscreenML out-performs methods such as pDockQ and iPTM for this task, and further that PPIscreenML exhibits impressive performance when identifying which ligand/receptor pairings engage one another across the structurally conserved tumor necrosis factor superfamily (TNFSF). Analysis of benchmark results using complexes not seen in PPIscreenML development strongly suggest that the model generalizes beyond training data, making it broadly applicable for identifying new protein complexes based on structural models built with AF2.