Protein-protein interactions (PPIs) are known to be involved in most cellular functions, and a detailed knowledge of such interactions is essential for studying their role in normal and pathological conditions. Significant progress is being made in the identification of PPIs through advances in computational methods. In particular, the AlphaFold2 machine learning-based model has been shown to accelerate drug discovery process by predicting the 3D structure of protein complexes. In this chapter, a straightforward protocol for predicting interprotein interactions between PAR-3 and its protein partner adapter molecule crk is provided. Such artificial intelligence-based and publicly available approaches can provide a resource for further investigation of therapeutic drug targets.

OBJECTIVE: Transient pregnancy-induced Cushing\'s syndrome is a rare condition characterized by the manifestation of symptoms solely during pregnancy, which typically resolve spontaneously following delivery or miscarriage. While it has been established that GNAS is associated with adrenal tumors, its specific role in the pathogenesis of pregnancy-induced Cushing\'s syndrome remains uncertain.This work aims to examine the association between GNAS mutation and pregnancy-induced Cushing\'s syndrome.
METHODS: DNA was extracted from patients\' peripheral blood and tumor tissues for whole-exome sequencing (WES) and Sanger sequencing. We used AlphaFold to predict the protein structure of wild-type and mutant GNAS and to make functional predictions, and immunohistochemistry was used to detect disease-associated protein expression. A review and summary of reported cases of transient pregnancy-induced Cushing\'s syndrome induced by pregnancy was conducted.
RESULTS: Using WES, we identified a somatic mutation in GNAS (NM_000516, c.C601T, p.R201C) that was predicted to have a deleterious effect using computational methods, such as AlphaFold. Human chorionic gonadotropin (hCG) stimulation tests had weakly positive results, and immunohistochemical staining of adrenal adenoma tissue also revealed positivity for luteinizing hormone/chorionic gonadotropin receptor (LHCGR) and cytochrome P450 family 11 subfamily B member 1 (CYP11B1). We reviewed 15 published cases of transient Cushing\'s syndrome induced by pregnancy. Among these cases, immunohistochemical staining of the adrenal gland showed positive LHCGR expression in 3 case reports, similar to our findings.
CONCLUSIONS: Transient pregnancy-induced Cushing\'s syndrome may be associated with somatic GNAS mutations and altered adrenal pathology due to abnormal activation of LHCGR.

A figure in the article by Barbarin-Bocahu & Graille [(2022), Acta Cryst. D78, 517-531] is corrected.

Neural network (NN)-based protein modeling methods have improved significantly in recent years. Although the overall accuracy of the two non-homology-based modeling methods, AlphaFold and RoseTTAFold, is outstanding, their performance for specific protein families has remained unexamined. G-protein-coupled receptor (GPCR) proteins are particularly interesting since they are involved in numerous pathways. This work directly compares the performance of these novel deep learning-based protein modeling methods for GPCRs with the most widely used template-based software-Modeller. We collected the experimentally determined structures of 73 GPCRs from the Protein Data Bank. The official AlphaFold repository and RoseTTAFold web service were used with default settings to predict five structures of each protein sequence. The predicted models were then aligned with the experimentally solved structures and evaluated by the root-mean-square deviation (RMSD) metric. If only looking at each program\'s top-scored structure, Modeller had the smallest average modeling RMSD of 2.17 Å, which is better than AlphaFold\'s 5.53 Å and RoseTTAFold\'s 6.28 Å, probably since Modeller already included many known structures as templates. However, the NN-based methods (AlphaFold and RoseTTAFold) outperformed Modeller in 21 and 15 out of the 73 cases with the top-scored model, respectively, where no good templates were available for Modeller. The larger RMSD values generated by the NN-based methods were primarily due to the differences in loop prediction compared to the crystal structures.

The breakthrough recently made in protein structure prediction by deep-learning programs such as AlphaFold and RoseTTAFold will certainly revolutionize biology over the coming decades. The scientific community is only starting to appreciate the various applications, benefits and limitations of these protein models. Yet, after the first thrills due to this revolution, it is important to evaluate the impact of the proposed models and their overall quality to avoid the misinterpretation or overinterpretation of these models by biologists. One of the first applications of these models is in solving the `phase problem\' encountered in X-ray crystallography in calculating electron-density maps from diffraction data. Indeed, the most frequently used technique to derive electron-density maps is molecular replacement. As this technique relies on knowledge of the structure of a protein that shares strong structural similarity with the studied protein, the availability of high-accuracy models is then definitely critical for successful structure solution. After the collection of a 2.45 Å resolution data set, we struggled for two years in trying to solve the crystal structure of a protein involved in the nonsense-mediated mRNA decay pathway, an mRNA quality-control pathway dedicated to the elimination of eukaryotic mRNAs harboring premature stop codons. We used different methods (isomorphous replacement, anomalous diffraction and molecular replacement) to determine this structure, but all failed until we straightforwardly succeeded thanks to both AlphaFold and RoseTTAFold models. Here, we describe how these new models helped us to solve this structure and conclude that in our case the AlphaFold model largely outcompetes the other models. We also discuss the importance of search-model generation for successful molecular replacement.

Artificial intelligence (AI)-based protein structure databases are expected to have an impact on drug discovery. Here, we show how AlphaFold could support rare diseases research programs. We focus on Alsin, a protein responsible for rare motor neuron diseases, such as infantile-onset ascending hereditary spastic paralysis (IAHSP) and juvenile primary lateral sclerosis (JPLS), and involved in some cases of amyotrophic lateral sclerosis (ALS). First, we compared the AlphaFoldDB human Alsin model with homology models of Alsin domains. We then evaluated the flexibility profile of Alsin and of experimentally characterized mutants present in patients with IAHSP. Next, we compared preliminary models of dimeric/tetrameric Alsin responsible for its physiological action with hypothetical models reported in the literature. Finally, we suggest the best animal model for drug candidates testing. Overall, we computationally show that drug discovery efforts toward Alsin-involving diseases should be pursued.