pi :: Psychiatry Investigation

Kim and Kim: Target Discovery Using Deep Learning-Based Molecular Docking and Predicted Protein Structures With AlphaFold for Novel Antipsychotics

Original Article

Psychiatry Investigation 2023;20(6):504-514.

Published online: May 30, 2023

DOI: https://doi.org/10.30773/pi.2022.0343

Target Discovery Using Deep Learning-Based Molecular Docking and Predicted Protein Structures With AlphaFold for Novel Antipsychotics

Yangsik Kim^1,²

, Seyong Kim²

¹Department of Psychiatry, Inha University Hospital, Incheon, Republic of Korea

²Mental Health Rearch Institute, National Center for Mental Health, Seoul, Republic of Korea

Correspondence: Yangsik Kim, MD, PhD Department of Psychiatry, Inha University Hospital, 27 Inhang-ro, Jung-gu, Incheon, Republic of Korea Tel: +82-32-890-3493, Fax: +82-32-890-3098, E-mail: ykim@inha.ac.kr

Received November 28, 2022 Revised January 31, 2023 Accepted March 6, 2023

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Objective

New drugs are needed to treat antipsychotic-resistant schizophrenia, especially those with clozapine-resistant schizophrenia. Atypical antipsychotics have predominantly 5-HT2A and dopaminergic antagonism, but also require investigation of other receptors.

Methods

In this study, the binding affinities between clozapine, olanzapine, and quetiapine with neuropharmacological, immunological, and metabolic receptors were measured using GNINA (Deep Learning Based Molecular Docking) and AlphaFold (Predicted Protein Structures).

Results

Through this study, it was determined that these antipsychotics showed high binding affinity to a variety of receptors, such as CB2, 5-HT1BR, NPYR4, and CCR5. Cyclosporin A and everolimus which show high affinities with those receptors could be used for the development of new antipsychotic drugs based on these drugs.

Conclusion

In the future, the method used in this study will be applied to the development of new antipsychotic drugs, including drug repositioning, and to the discovery of the pathophysiology of schizophrenia.

Keywords: Schizophrenia, Machine learning, Drug discovery, Clozapine, Antipsychotic

INTRODUCTION

Schizophrenia is a chronic mental illness accompanied by psychotic symptoms such as hallucination and delusion, negative symptoms such as avolition and affective blunting, and cognitive impairment. This condition is the eleventh leading cause of years lived with disability in 2013 [1,2]. Drugs for psychiatric disorders, including schizophrenia, mainly act on monoamine receptors, and have an overall therapeutic effect of approximately 50%. Intractable cases occur in about 20% of patients [3,4]. Factors that include low drug adherence, as well as psychiatric and medical comorbidities, contribute to drug treatment resistance. However, antipsychotics resistance appears even in schizophrenia patients with high adherence to treatment and no comorbidities, which results in unmet medical demand [5].

Clozapine is an antipsychotic drug that has the best effect on auditory hallucinations and delusions among antipsychotics, and has fewer extrapyramidal side effects such as the Parkinsonian-like behavior induced by antipsychotic drugs [6]. Clozapine was shown to be effective in about 70% of patients who were resistant to antipsychotic drugs other than clozapine. However, in a previous study, about 50% of schizophrenia symptoms persisted for more than 12 years among clozapine-treated patients and it is estimated that clozapine treatment resistance reaches 10% of patients with schizophrenia [7]. Since clozapine began to be used to treat schizophrenia patients in the late 1980s, antipsychotic drugs that show better effect than clozapine have not been developed. This is due to the lack of understanding of the pathophysiology of mental disorders, the incomplete identification of molecular-pharmacological targets of existing drugs, the lack of preclinical models, and the factors that make it difficult to predict accurately the clinical response due to the diversity of disease courses [8-10]. To improve the symptoms of clozapine-resistant schizophrenia patients, additional pharmacological or biological treatments such as antidepressants, mood stabilizers, and electroconvulsive therapy are combined, but the effect is just temporary or insufficient to control the mental illness. Thus, a new treatment target and related drug discovery is needed to overcome schizophrenia.

Olanzapine and quetiapine are classified as thienobenzodiazepine and dibenzothiazepine, respectively. They have chemical structures similar to clozapine, which is classified as dibenzodiazepine (Figure 1) [11]. Quetiapine is used in the guideline as a first-line drug for schizophrenia and bipolar affective disorder. Olanzapine has stronger efficacy but also a strong weight-gain effect, so it is used as a second-line drug [12-14]. Like most small molecule drugs, clozapine, olanzapine, and quetiapine all have multiple targets. In particular, psychiatric drugs act as magic shotguns rather than magic bullets [9].

Known targets of clozapine include 5-HT2A, 5-HT2C, D2, D4, alpha1, alpha2, H1, M1-5, N-methyl-D-aspartate (NMDA), and GABAB receptors [15,16]. Drugs such as Ulotaront, the TAAR1 agonistic and non-D2 antagonistic antipsychotic offers hope to treat schizophrenia [17,18]. However, overall, there are currently no antipsychotics for schizophrenia more effective than clozapine, and new antipsychotic drugs for clozapine-resistant patients are urgently needed.

In this study, GNINA, a pretrained convolutional neural network (CNN) scoring function, was used to calculate binding poses. It is based on learned representations of 3D proteinligand interactions. GNINA, a fork of Smina, outperforms Smina and AutoDock Vina in virtual screening methods [19,20]. Also in this study, the protein structure predicted by AlphaFold2 was used, especially the drug target receptors [21]. Through this study, receptors known to affect schizophrenia-related neurotransmitters, as well as immunological, metabolic, and neural stem cell growth were searched and found in the AlphaFold Protein Structure Database. Then they were analyzed with GNINA. The differences in receptor affinities according to drugs were measured, and receptors that could be used in drug discovery for schizophrenia are proposed for further study.

In detail, using the already developed GNINA and AlphaFold, the neuropharmacological, immune, and metabolic receptors that specifically act on each drug among clozapine, olanzapine, and quetiapine are explored. And based on these results, they discovered specific drugs that act on these receptors.

METHODS

Supplementary Table 1 (in the online-only Data Supplement) is a list of the receptors included in this study, which were classified as neuropharmacological (neuro), immunological (immune), or metabolic receptors [8,22-27]. In particular, it contains receptors involved in the metabolism of neural stem cells [27-30].

Ligands other than clozapine, olanzapine, and quetiapine in Supplementary Table 2 (in the online-only Data Supplement) were included in this study as the most commonly prescribed drugs [31-33] (including illegal drugs) and drugs used for targeted therapy [34]. Monoclonal antibodies were excluded because they could not be measured using GNINA.

The protein pdb file of the target receptor was downloaded from the AlphaFold Protein Structure Database (https://alphafold.ebi.ac.uk/). The downloading code was automated using the crawling library called selenium and request.

PDBFixer in OpenMM was used to perform the tasks that needed to be processed before performing docking simulation between protein and ligand [35]. In short, PDBFixer is an application that fixes problems in the Protein Data Bank files to prepare them for simulation.

DeepChem (https://deepchem.io/) makes the most suitable pose for protein and ligand docking. This prepares protein-ligand complexes for docking. Afterwards, using the processed pdb file and SMILES of the drug, we declared PoseGenerator in GNINA, and docked the protein and ligand with generate_poses in the default scoring setting. As a result, scores displayed the results obtained through a pretrained CNN, including binding affinity. The scores include binding affinities (kcal/mol), CNN pose scores (a probability that the pose has a low root mean square deviation to the binding pose), and CNN affinities predicted by GNINA19. The Pearson correlation efficient value between binding affinities and CNN affinities was -0.66, (p<0.0001). In this study, the results were compared using only CNN affinity values. We used the highest values of CNN affinities, and there was no difference in the order of receptors between the lowest and highest values.

The PyMOL Molecular Graphics System, Version 1.8 (Schrödinger LLC., New York, USA) was used to visualize the ligand-target interaction. No institutional review board approval is required as no humans or animals were involved in this study.

RESULTS

Top 20 CNN affinities by receptor

We obtained CNN binding affinities values for each of the 220 receptors that interacted with the three drugs via GNINA (Figure 2). The highest CNN affinities value was obtained from clozapine-bound cannabinoid receptor 2 (CB2, 7.522), olanzapine-bound serotonin 1B receptor (5-HT1BR, 7.004), olanzapine-bound muscarinic cholinergic receptor 5 receptor (M5, 6.713), and olanzapine-bound C-C chemokine receptor type 5 receptor (CCR5, 6.677) followed. Moreover, quetiapine showed the highest value with neuropeptide Y 4 receptor (NPYR4, 6.657). Overall, out of the top 20, 8 neuro receptors (red), 5 immune receptors (green), and 7 metabolic (blue) receptors were included in the list. Figure 2 shows the 3D structure of the receptor with the highest CNN affinities for each ligand.

CNN affinities among receptors for antipsychotics

Among the receptors bound to clozapine, CB2 had the highest value, followed by CCR4 (6.511), CCR5 (6.500), CCR3 (6.451), and C-X-C chemokine receptor 3 (CXCR3, 6.424). 4 neuro receptors, 10 immune receptors, and 6 metabolic receptors were included in the list for CNN affinities among receptors for clozapine, and immune receptors were the most common in the top 20. Keratin was included in the analysis as a control for the receptors.

Among the receptors bound to olanzapine, 5-HT1BR had the highest value, followed by M5 (6.713), CCR5 (6.677), M1 (6.534), and CCR3 (6.459). Neuro receptors were 5, immune receptors 4, metabolic receptors 11 in the list, and metabolic receptors were the most common.

Among the receptors bound to quetiapine, NPYR4 had the highest value, followed by fibroblast growth factor receptor 3 (FGFR3, 6.436), CCR5 (6.296), CCR8 (6.239), and patched domain containing 4 (PTCHD4, 6.234). Neuro receptors were 8, immune receptors 6, metabolic receptors 6 in the list, and metabolic receptors were the most common (Table 1).

CNN affinities among the receptors bound by the classified receptor system

In neuro receptors, the order of CNN affinities was CB2 (7.522), 5-HT1BR (7.004), M5 (6.713), M1 (6.534), and 5-HT2AR (6.309). More than two receptors related to 5-HT, muscarinic, CB, and melatonin were included, and TAAR, VIAAT, VGLUT1, and GRIK4 were also included.

Among immune receptors, CNN affinities of CCR5 (6.677), CCR4 (6.511), CCR3 (6.459), CXCR3 (6.424), and CCR8 (6.239) were shown. Two or more CCR, CXCR, and prostaglandin-related proteins were included, and the C5a receptor, toll-like receptor 2, and prostacyclin receptor were also included.

In the metabolic receptors, CNN affinities were shown in the order of NPYR4 (6.657), FGFR3 (6.436), vascular endothelial growth factor 3 (VEGFR3, 6.368), NPYR1 (6.329), and PTCHD4 (6.234). Two or more NPYR, orexin receptor, FGFR and melanocortin receptor-related proteins were included, and vasopressin receptor, oxytocin receptor, progesterone receptor, platelet-derived growth factor, Tropomyosin receptor kinase C receptor, and glial cell-derived neurotrophic factor were also included (Table 2).

Top 20 CNN affinities among widely used drugs and the highest binding receptors for antipsychotics

We found that CB2, 5-HT1BR, NPYR4, and CCR5 showed high affinities with the antipsychotics through this study, so we performed additional analysis to search for affinities with a wider range of ligands for these receptors.

Surprisingly, cyclosporine showed the highest affinities for all three receptors except for CCR5 (CB2 7.996, 5-HT1BR 7.423, NPYR4 7,785), everolimus had the highest affinities for CCR5 (7.584), and the other three receptors had the second highest affinities with it (CB2 7.388, 5-HT1BR 7.422, NPYR4 7.420).

In addition, for CB2: octreotide (7.120), trametinib (7.036), and maraviroc (6.979) showed high affinities (in this order), and for 5-HT1BR: salinomycin (7.105), tacrolimus (7.050), nilotinib (6.913), and for NPYR4: crizotinib (7.100), nilotinib (6.976), lapatinib (6.953), and for CCR5: nilotinib (7.420), crizotinib (7.417), maraviroc (7.410), and cyclosporine (7.373) were found in that order (Figure 3).

DISCUSSION

Throughout this study, we used CNN-based GNINA to estimate the affinities of clozapine, olanzapine, and quetiapine with neuropharmacological, immunological, and metabolic receptors predicted from AlphaFold. Through this in silico study, we confirmed that antipsychotic drugs act not only on neuropharmacological receptors but also on immune and metabolic receptors. In particular, CB2 for clozapine, 5-HT1BR for olanzapine, and NPYR4 for quetiapine showed the highest affinities.

Neurons and glia in the brain are expected to account for about 100 billion cells, respectively [36]. In addition to the rapid electrophysiological properties of neurons, which are excitable cells, the modulatory effect of G protein-coupled receptor is also important for cellular functions, including signal transduction [37,38]. The function of neurons cannot be maintained without the role of the glia cells that make up the brain. Therefore, it is essential to understand the action of antipsychotics on immunological and metabolic receptors as well as neuropharmacological receptors. In this study, we performed an exploratory study involving a variety of receptors to understand the interactions of neurons and glia and the functioning of the brain within the body’s immune and metabolic environment.

In the field of psychiatry, drug discovery including virtual screening is actively progressing, but there are still unmet needs and it is one of the most difficult areas to develop new drugs [39,40]. However, looking at some studies using deep learning and virtual screening, there is a study that predict drug repositioning using various deep learning techniques in psychiatric disorder including schizophrenia [41]. In addition, Another study utilized deep learning techniques to structurally predict the GABA receptor action of herbal compounds [42]. Finally, one study conducted to discover optimized NMDA receptor antagonists for Alzheimer’s disease using Smina and idock for virtual screening [43].

In a previous study of CB2, nabilone-induced exacerbation of psychotic symptoms was reported in patients with Parkinson’s disease [44]. Nabilone is a synthetic form of tetrahydrocannabinol for nausea and vomiting in cancer patients and HIV patients; it acts as a partial agonist for both CB1 and CB2. The most worrisome side effect is psychotic symptoms, but only two cases have been reported in studies targeting post-traumatic stress disorder patients [45]. Unlike CB1, which is distributed in the central nervous system, CB2 is distributed in the peripheral nervous system and also in immune cells such as lymphoid tissue and microglia [46]. The antibiotic minocycline acts on CB2 and is expected to serve as an adjuvant treatment to prevent the onset or exacerbation of schizophrenia [47-49].

Previous studies on 5-HT1BR have shown that triptan drugs, such as sumatriptan for migraine treatment, act on receptors such as 5-HT1AR, 5-HT1DR, 5-HT1FR, and 5-HT1BR [50]. The action of 5-HT1BR as a selective serotonin reuptake inhibitor is also reported [51]. Increased 5-HT1BR messenger ribonucleic acid in the dorsolateral prefronal cortex tissue [52], and an increase in 5-HT1BR mRNA in the hippocampus were also reported in a schizophrenia biomarker study [53].

A previous study on NPYR4 showed that obinepitide was an investigational drug acting on NPYR4, and it showed weight loss in experimental animals by acting on NPYR2 and NPYR4 [54]. Because of the high comorbidity of weight gain and metabolic syndrome due to schizophrenia itself and antipsychotics, it seems to have both the implications of a therapeutic target and a receptor associated with drug side effects [55]. NPYR4 has not been studied for schizophrenia, thus this study suggests a new receptor target.

CCR5 is a chemokine receptor for regulated on activation, normal T cell expressed and secreted as its ligand and is known to be expressed in glia and neural stem cells [56]. Because CCR5 is known as the infection route of HIV 1, maraviroc (which inhibits CCR5) has been developed and used [57]. Moreover, there is no study about treating schizophrenia using maraviroc. In a CCR5 gene deletion animal study, CCR5-deleted mice showed a decrease in social recognition [58]. In a schizophrenia genome study, higher expression of CCR5 was reported compared to the control [59], and the frequencies of CCR5-A55029GA genotypes and CCR5-A55029GAG genotypes were found to be higher in patients. In a genomic study of 268 schizophrenia and 323 controls, association with the CCR5 32 deletion allele was reported in the late onset schizophrenia group [60]. After combining previous studies, it is now expected that the regulation of CCR5 may affect the treatment of patients with schizophrenia.

Cyclosporine is a calcineurin inhibitor mainly used as an immunosuppressant. Cyclosporine-induced neurotoxicity is a well-known side effect and, in particular, can induce psychosis [61-65]. Everolimus, also an immunosuppressant, is a drug that acts on the mechanistic target of rapamycin and that affects the growth of neural stem cells [66]. A case study has been reported in which everolimus also induced psychosis [67], but in a previous study, the use of everolimus in children and adolescents with tuberous sclerosis resulted in a decrease in prepulse inhibition, an indicator of psychosis and sensorimotor gating [68-70]. In this study, high binding affinities between cyclosporine and everolimus, and the receptors found in this study, were discovered. Now, a study to develop drugs that bind to the corresponding receptors, but have the opposite action through pharmacodynamics, can be considered.

In this study, we evaluated only CNN affinity by ligand-receptor binding in silico, and evaluation of the pharmacodynamic effect for function is needed in the future. Moreover, physical measurements of ligand-receptor binding, such as provided by x-ray crystallography, are also required for a new antipsychotics discovery. Using CNN-based GNINA, we could perform a more accurate deep learning-based analysis than with the existing AutoDock Vina. In addition, a wide range of receptors were searched, and in particular, the limitations of existing protein structure studies were supplemented using AlphaFold. By discovering a target for new antipsychotics that had not been studied before (such as NPYR4), we got an opportunity to provide new insight into the field of drug discovery for schizophrenia.

This study investigated olanzapine and quetiapine, which have chemical structures similar to those of clozapine, and which are the most effective drugs for controlling psychotic symptoms of schizophrenia. Furthermore, receptors upon which each drug acts were searched using GNINA for structure-based drug discovery (SBDD). It was confirmed that antipsychotics have high affinities on receptors (such as CB2 and CCR5) that are present in immune cells such as microglia, which exist in the immune part of the central nervous system. The receptors described in this study are suggested to play important roles in the improvement of psychiatric symptoms and the prevention of recurrence, and furthermore, in the pathophysiology of schizophrenia. Cyclosporin A and everolimus, which bind strongly to these receptors, have potential applications in the development of novel antipsychotic drugs.

Clozapine, olanzapine, and quetiapine have been studied mainly on serotonin and dopamine receptors in previous studies because of limitations in biological resources available in the past (structure of receptor proteins, accessibility, measurable technology, etc.) [71]. And there is a high need to explore the mechanism of drugs on a wider range of targets including receptor proteins. In addition, the relationship between immunological and metabolic abnormalities and psychosis was recently revealed [72,73]. Therefore, the effect on immunological and metabolic receptors could show therapeutic effects, and finding new receptors with new mechanisms and drugs specific to those receptors is expected to lead to the discovery of treatments for antipsychotic-resistant schizophrenia patients.

Although the mechanism of mental illness is still unknown, the therapeutic effect of antipsychotic drugs is obvious. Therefore, it is considered as a useful approach to explore the receptor binding of antipsychotic drugs and use them to search for new targets. We hope that the method used in this study for SBDD and drug repositioning for the development of new antipsychotics will be of great help to schizophrenia patients.

Supplementary Materials

The online-only Data Supplement is available with this article at https://doi.org/10.30773/pi.2022.0343.

Supplementary Table 1.

List of receptors classified by system

pi-2022-0343-Supplementary-Table-1.pdf

Supplementary Table 2.

List of ligands: the most commonly prescribed drugs and drugs used for targeted therapy

pi-2022-0343-Supplementary-Table-2.pdf

Notes

Availability of Data and Material

All data generated or analyzed during the study are included in this published article (and its supplementary information files).

Conflicts of Interest

The authors have no potential conflicts of interest to disclose.

Author Contributions

Conceptualization: Yangsik Kim. Formal analysis: Yangsik Kim. Funding acquisition: Yangsik Kim. Methodology: Seyong Kim. Resources: Seyong Kim. Software: Seyong Kim. Supervision: Yangsik Kim. Visualization: Yangsik Kim. Writing—original draft: Yangsik Kim. Writing—review & editing: Yangsik Kim.

Funding Statement

This work was supported by a National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (NRF-2021R1C1C1003266 to Y.K.), and the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (HI22C0492 to Y.K.).

Figure 1.

Chemical structures of antipsychotic and immunosuppressive drugs focused in deep learning-based molecular docking. A: Clozapine (the compound number in PubChem, 135398737). B: Olanzapine (135398745). C: Quetiapine (5002). D: Cyclosporin A (5284373). E: Everolimus (6442177).

Figure 2.

Top 20 convolutional neural network (CNN) affinities by receptor. A: A representative image of binding between cannabinoid receptor 2 (CB2) and clozapine. B: A representative image of binding between sertotonin receptor 1B (5-HT1BR) and olanzapine. C: A representative image of binding between neuropeptide Y receptor 4 (NPYR4) and quetiapine. D: A representative image of binding between C-C chemokine receptor 5 (CCR5) and olanzapine. E: Top 20 list of CNN affinities by receptor, neuropharmacological receptors are shown in red, immunological receptors in green, and metabolic receptors in blue. Shading in the table indicates the drug-receptor binding pair with the highest affinity. 5-HT, 5-hydroxytryptamine (serotonin) receptor; CB, cannabinoid receptor; CCR, C-C chemokine receptor; CXCR, C-X-C chemokine receptor; FGFR, fibroblast growth factor receptor; M, Muscarinic acetylcholine receptor; NPYR, neuropeptide Y receptor; PTCHD, patched domain-containing protein; VIAAT, vesicular inhibitory amino acid transporter.

Figure 3.

Top 20 convolutional neural network (CNN) affinities among widely used drugs and the highest binding receptors for antipsychotics. A: A representative image of binding between cannabinoid receptor 2 (CB2) and cyclosporin A. B: A representative image of binding between CB2 and everolimus. C: A representative image of binding between sertotonin receptor 1B (5-HT1BR) and cyclosporin A. D: A representative image of binding between 5-HT1BR and everolimus. E: A representative image of binding between neuropeptide Y receptor 4 (NPYR4) and cyclosporin A. F: A representative image of binding between NPYR4 and everolimus. G: A representative image of binding between C-C chemokine receptor 5 (CCR5) and cyclosporin A. H: A representative image of binding between CCR5 and everolimus. I: Top 20 list of CNN affinities by receptors found in this study including CB2, 5-HT1BR, NPYR4, and CCR5, neuropharmacological receptors are shown in red, immunological receptors in green, and metabolic receptors in blue. LSD, lysergic acid diethylamide; THC, tetrahydrocannabinol.

Table 1.

Top 20 CNN affinities among receptors for antipsychotics

	CNN pose	CNN affinities
Clozapine
CB2	0.7190	7.522
CCR4	0.7282	6.511
CCR5	0.8832	6.500
CCR3	0.6212	6.451
CXCR3	0.6121	6.424
NPYR1	0.8153	6.329
5-HT2AR	0.9436	6.309
VIAAT	0.8697	6.298
M1	0.7386	6.241
CCR8	0.8230	6.200
NPYR5	0.6819	6.156
OrexinR1	0.7103	6.086
PTCHD4	0.6221	6.082
CXCR6	0.5169	6.074
NPYR2	0.6822	6.065
CXCR5	0.6991	5.932
CXCR2	0.8291	5.918
CCR7	0.8554	5.914
PDGFb	0.5022	5.905
ProstaglandinE2REP4	0.7796	5.893
Keratin	0.8034	4.764
Olanzapine
5-HT1BR	0.7759	7.004
M5	0.6796	6.713
CCR5	0.6367	6.677
M1	0.7347	6.534
CCR3	0.6083	6.459
VEGFR3	0.5331	6.368
CCR4	0.7912	6.363
M4	0.5657	6.241
NPYR4	0.5697	6.185
FGFR4	0.5648	6.178
NPYR5	0.5715	6.178
NPYR1	0.6426	6.113
NPYR2	0.7241	6.085
PTCHD1	0.7769	6.077
CXCR5	0.5971	5.998
FGFR3	0.3654	5.954
IRRP	0.5311	5.910
OrexinR1	0.6282	5.825
GABA epsilon	0.6251	5.810
OrexinR2	0.6036	5.806
Keratin	0.7428	4.529
Quetiapine
NPYR4	0.8162	6.657
FGFR3	0.5511	6.436
CCR5	0.5938	6.296
CCR8	0.5906	6.239
PTCHD4	0.7645	6.234
MelatoninR1b	0.9085	6.153
M4	0.5366	6.147
CCR7	0.9301	6.028
NPYR1	0.5912	6.026
CXCR5	0.5576	6.024
CB1	0.9257	6.024
V2	0.9054	6.011
MelatoninR1a	0.6645	5.996
NPYR5	0.4089	5.985
TAAR1	0.7617	5.984
5-HT2AR	0.8096	5.981
ProstaglandinE2REP4	0.7903	5.981
5-HT2CR	0.5341	5.976
GRIK4	0.8778	5.956
CXCR3	0.4381	5.953
Keratin	0.6079	4.218

CNN, convolutional neural network; 5-HT, 5-hydroxytryptamine (serotonin) receptor; CB, cannabinoid receptor; CC, C-C chemokine receptor; CXC, C-X-C chemokine receptor; FGFR, fibroblast growth factor receptor; GABA, Gamma-aminobutyric acid receptor; GRIK, Glutamate receptor ionotropic, kainite; IRRP, insulin receptor related protine; VEGFR, vascular endothelial growth factor receptor; VIAAT, vesicular inhibitory amino acid transporter; V, vassopressin receptor

Table 2.

Top 20 CNN affinities among receptors bound by classified receptor system

	CNN pose	CNN affinities	Olan	Quet	Clo
Neuro
CB2	0.7190	7.522			^*
5-HT1BR	0.7759	7.004	^*
M5	0.6796	6.713	^*
M1	0.7347	6.534	^*
5-HT2AR	0.9436	6.309			^*
VIAAT	0.8697	6.298			^*
M4	0.5657	6.241	^*
MelatoninR1b	0.9085	6.153		^*
CB1	0.9257	6.024		^*
MelatoninR1a	0.6645	5.996		^*
TAAR1	0.7617	5.984		^*
5-HT2CR	0.5341	5.976		^*
GRIK4	0.8778	5.956		^*
5-HT6R	0.8680	5.950		^*
VGLUT1	0.7824	5.916		^*
5-HT3AR	0.9049	5.874		^*
HistamineR1	0.7191	5.857		^*
5-HT1AR	0.5862	5.856		^*
GABA epsilon	0.4967	5.847		^*
GABA rho-1	0.7752	5.816			^*
Immune
CCR5	0.6367	6.677	^*
CCR4	0.7282	6.511			^*
CCR3	0.6083	6.459	^*
CXCR3	0.6121	6.424			^*
CCR8	0.5906	6.239		^*
CXCR6	0.5169	6.074			^*
CCR7	0.9301	6.028		^*
CXCR5	0.5576	6.024		^*
ProstaglandinE2REP4	0.7903	5.981		^*
CXCR1	0.4841	5.944		^*
CXCR2	0.8291	5.918			^*
ProstaglandinD2R	0.6284	5.815		^*
ProstaglandinF2aR	0.7264	5.776		^*
ProstacyclinR	0.5927	5.664		^*
C5aR	0.5092	5.646		^*
ProstaglandinE2REP1	0.6847	5.614		^*
CXCR4	0.6879	5.600			^*
CCR6	0.6100	5.600			^*
ProstaglandinD2R2	0.8643	5.597		^*
TLR2	0.7450	5.589		^*
Metabolic
NPYR4	0.8162	6.657		^*
FGFR3	0.5511	6.436		^*
VEGFR3	0.5331	6.368	^*
NPYR1	0.8153	6.329			^*
PTCHD4	0.7645	6.234		^*
FGFR4	0.5648	6.178	^*
NPYR5	0.5715	6.178	^*
OrexinR1	0.7103	6.086			^*
NPYR2	0.7241	6.085	^*
PTCHD1	0.7769	6.077	^*
V2	0.9054	6.011		^*
IRRP	0.5311	5.910	^*
PDGFb	0.5022	5.905			^*
TrkC	0.6214	5.877			^*
MCR4	0.8809	5.826			^*
Putative NPYR6	0.7462	5.824		^*
NPYR4-2	0.8125	5.810		^*
GDNFR1	0.5613	5.807		^*
OrexinR2	0.6036	5.806	^*
OxytocinR	0.6559	5.713			^*

^* the drug-receptor binding pair with the highest affinity.

CNN, convolutional neural network; 5-HT, 5-hydroxytryptamine (serotonin) receptor; CB, cannabinoid receptor; CC, C-C chemokine receptor; CXC, C-X-C chemokine receptor; FGFR, fibroblast growth factor receptor; GABA, gamma-aminobutyric acid receptor; GRIK, glutamate receptor ionotropic kainite; IRRP, insulin receptor related protine; MCR, melanocortin receptor; TAAR, trace amino acid receptor; VEGFR, vascular endothelial growth factor receptor; VIAAT, vesicular inhibitory amino acid transporter; V, vassopressin receptor

REFERENCES

1. Vigo D, Thornicroft G, Atun R. Estimating the true global burden of mental illness. Lancet Psychiatry 2016;3:171–178.

2. Charlson FJ, Ferrari AJ, Santomauro DF, Diminic S, Stockings E, Scott JG, et al. Global epidemiology and burden of schizophrenia: findings from the global burden of disease study 2016. Schizophr Bull 2018;44:1195–1203.

3. Hyman SE. Psychiatric drug development: diagnosing a crisis. Cerebrum 2013;2013:5

4. Huhn M, Tardy M, Spineli LM, Kissling W, Förstl H, Pitschel-Walz G, et al. Efficacy of pharmacotherapy and psychotherapy for adult psychiatric disorders: a systematic overview of meta-analyses. JAMA Psychiatry 2014;71:706–715.

5. Conley RR, Kelly DL. Management of treatment resistance in schizophrenia. Biol Psychiatry 2001;50:898–911.

6. Nielsen J, Damkier P, Lublin H, Taylor D. Optimizing clozapine treatment. Acta Psychiatr Scand 2011;123:411–422.

7. Juul Povlsen U, Noring U, Fog R, Gerlach J. Tolerability and therapeutic effect of clozapine. A retrospective investigation of 216 patients treated with clozapine for up to 12 years. Acta Psychiatr Scand 1985;71:176–185.

8. Pratt J, Winchester C, Dawson N, Morris B. Advancing schizophrenia drug discovery: optimizing rodent models to bridge the translational gap. Nat Rev Drug Discov 2012;11:560–579.

9. Roth BL, Sheffler DJ, Kroeze WK. Magic shotguns versus magic bullets: selectively non-selective drugs for mood disorders and schizophrenia. Nat Rev Drug Discov 2004;3:353–359.

10. Javitt DC, Spencer KM, Thaker GK, Winterer G, Hajós M. Neurophysiological biomarkers for drug development in schizophrenia. Nat Rev Drug Discov 2008;7:68–83.

11. Jafari S, Fernandez-Enright F, Huang XF. Structural contributions of antipsychotic drugs to their therapeutic profiles and metabolic side effects. J Neurochem 2012;120:371–384.

12. Keepers GA, Fochtmann LJ, Anzia JM, Benjamin S, Lyness JM, Mojtabai R, et al. The American Psychiatric Association practice guideline for the treatment of patients with schizophrenia. Am J Psychiatry 2020;177:868–872.

13. Verdolini N, Hidalgo-Mazzei D, Del Matto L, Muscas M, Pacchiarotti I, Murru A, et al. Long-term treatment of bipolar disorder type I: a systematic and critical review of clinical guidelines with derived practice algorithms. Bipolar Disord 2021;23:324–340.

14. Lieberman JA, Stroup TS, McEvoy JP, Swartz MS, Rosenheck RA, Perkins DO, et al. Effectiveness of antipsychotic drugs in patients with chronic schizophrenia. N Engl J Med 2005;353:1209–1223.

15. Nair PC, McKinnon RA, Miners JO, Bastiampillai T. Binding of clozapine to the GABAB receptor: clinical and structural insights. Mol Psychiatry 2020;25:1910–1919.

16. Khokhar JY, Henricks AM, Sullivan EDK, Green AI. Unique effects of clozapine: a pharmacological perspective. Adv Pharmacol 2018;82:137–162.

17. Revel FG, Moreau JL, Pouzet B, Mory R, Bradaia A, Buchy D, et al. A new perspective for schizophrenia: TAAR1 agonists reveal antipsychotic- and antidepressant-like activity, improve cognition and control body weight. Mol Psychiatry 2013;18:543–556.

18. Koblan KS, Kent J, Hopkins SC, Krystal JH, Cheng H, Goldman R, et al. A Non-D2-receptor-binding drug for the treatment of schizophrenia. N Engl J Med 2020;382:1497–1506.

19. McNutt AT, Francoeur P, Aggarwal R, Masuda T, Meli R, Ragoza M, et al. GNINA 1.0: molecular docking with deep learning. J Cheminform 2021;13:43

20. Sunseri J, Koes DR. Virtual screening with Gnina 1.0. Molecules 2021;26:7369

21. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature 2021;596:583–589.

22. Agid Y, Buzsáki G, Diamond DM, Frackowiak R, Giedd J, Girault JA, et al. How can drug discovery for psychiatric disorders be improved? Nat Rev Drug Discov 2007;6:189–201.

23. Dwyer DS, Weeks K, Aamodt EJ. Drug discovery based on genetic and metabolic findings in schizophrenia. Expert Rev Clin Pharmacol 2008;1:773–789.

24. Leonard BE. Is there an immunologic basis for schizophrenia? Expert Rev Clin Immunol 2005;1:103–112.

25. Readhead B, Hartley BJ, Eastwood BJ, Collier DA, Evans D, Farias R, et al. Expression-based drug screening of neural progenitor cells from individuals with schizophrenia. Nat Commun 2018;9:4412

26. Collier DA, Eastwood BJ, Malki K, Mokrab Y. Advances in the genetics of schizophrenia: toward a network and pathway view for drug discovery. Ann N Y Acad Sci 2016;1366:61–75.

27. Reif A, Fritzen S, Finger M, Strobel A, Lauer M, Schmitt A, et al. Neural stem cell proliferation is decreased in schizophrenia, but not in depression. Mol Psychiatry 2006;11:514–522.

28. Gorba T, Conti L. Neural stem cells as tools for drug discovery: novel platforms and approaches. Expert Opin Drug Discov 2013;8:1083–1094.

29. Taupin P. Therapeutic potential of adult neural stem cells. Recent Pat CNS Drug Discov 2006;1:299–303.

30. Rubin LL. Stem cells and drug discovery: the beginning of a new era? Cell 2008;132:549–552.

31. Smith DA, Di L, Kerns EH. The effect of plasma protein binding on in vivo efficacy: misconceptions in drug discovery. Nat Rev Drug Discov 2010;9:929–939.

32. Fuentes AV, Pineda MD, Venkata KCN. Comprehension of top 200 prescribed drugs in the US as a resource for pharmacy teaching, training and practice. Pharmacy (Basel) 2018;6:43

33. Gan J, Ruan Q, He B, Zhu M, Shyu WC, Humphreys WG. In vitro screening of 50 highly prescribed drugs for thiol adduct formation--comparison of potential for drug-induced toxicity and extent of adduct formation. Chem Res Toxicol 2009;22:690–698.

34. Murdoch D, Sager J. Will targeted therapy hold its promise? An evidence-based review. Curr Opin Oncol 2008;20:104–111.

35. Eastman P, Swails J, Chodera JD, McGibbon RT, Zhao Y, Beauchamp KA, et al. OpenMM 7: rapid development of high performance algorithms for molecular dynamics. PLoS Comput Biol 2017;13:e1005659.

36. Herculano-Houzel S. The glia/neuron ratio: how it varies uniformly across brain structures and species and what that means for brain physiology and evolution. Glia 2014;62:1377–1391.

37. Turrigiano GG, Nelson SB. Homeostatic plasticity in the developing nervous system. Nat Rev Neurosci 2004;5:97–107.

38. Keck T, Toyoizumi T, Chen L, Doiron B, Feldman DE, Fox K, et al. Integrating Hebbian and homeostatic plasticity: the current state of the field and future research directions. Philos Trans R Soc Lond B Biol Sci 2017;372:20160158

39. Berk M, Nierenberg AA. Three paths to drug discovery in psychiatry. Am J Psychiatry 2015;172:412–414.

40. Vatansever S, Schlessinger A, Wacker D, Kaniskan HÜ, Jin J, Zhou MM, et al. Artificial intelligence and machine learning-aided drug discovery in central nervous system diseases: state-of-the-arts and future directions. Med Res Rev 2021;41:1427–1473.

41. Zhao K, So HC. Drug repositioning for schizophrenia and depression/anxiety disorders: a machine learning approach leveraging expression data. IEEE J Biomed Health Inform 2018;23:1304–1315.

42. Marunnan SM, Pulikkal BP, Jabamalairaj A, Bandaru S, Yadav M, Nayarisseri A, et al. Development of MLR and SVM aided QSAR models to identify common SAR of GABA uptake herbal inhibitors used in the treatment of schizophrenia. Curr Neuropharmacol 2017;15:1085–1092.

43. Sharma G, Shukla R, Singh TR. Identification of small molecules against the NMDAR: an insight from virtual screening, density functional theory, free energy landscape and molecular dynamics simulation-based findings. Netw Model Anal Health Inform Bioinform 2022;11:31

44. Udow SJ, Freitas ME, Fox SH, Lang AE. Exacerbation of psychosis triggered by a synthetic cannabinoid in a 70-year-old woman with Parkinson disease. CMAJ 2018;190:E50–E52.

45. Cameron C, Watson D, Robinson J. Use of a synthetic cannabinoid in a correctional population for posttraumatic stress disorder-related insomnia and nightmares, chronic pain, harm reduction, and other indications: a retrospective evaluation. J Clin Psychopharmacol 2014;34:559–564.

46. Mackie K. Cannabinoid receptors as therapeutic targets. Annu Rev Pharmacol Toxicol 2006;46:101–122.

47. Solmi M, Veronese N, Thapa N, Facchini S, Stubbs B, Fornaro M, et al. Systematic review and meta-analysis of the efficacy and safety of minocycline in schizophrenia. CNS Spectr 2017;22:415–426.

48. Liu F, Guo X, Wu R, Ou J, Zheng Y, Zhang B, et al. Minocycline supplementation for treatment of negative symptoms in early-phase schizophrenia: a double blind, randomized, controlled trial. Schizophr Res 2014;153:169–176.

49. Cortez IL, Rodrigues da Silva N, Guimarães FS, Gomes FV. Are CB2 receptors a new target for schizophrenia treatment? Front Psychiatry 2020;11:587154

50. Deen M, Hougaard A, Hansen HD, Schain M, Dyssegaard A, Knudsen GM, et al. Association between sumatriptan treatment during a migraine attack and central 5-HT1B receptor binding. JAMA Neurol 2019;76:834–840.

51. Ruf BM, Bhagwagar Z. The 5-HT1B receptor: a novel target for the pathophysiology of depression. Curr Drug Targets 2009;10:1118–1138.

52. López-Figueroa AL, Norton CS, López-Figueroa MO, Armellini-Dodel D, Burke S, Akil H, et al. Serotonin 5-HT1A, 5-HT1B, and 5-HT2A receptor mRNA expression in subjects with major depression, bipolar disorder, and schizophrenia. Biol Psychiatry 2004;55:225–233.

53. Mohammadi A, Rashidi E, Amooeian VG. Brain, blood, cerebrospinal fluid, and serum biomarkers in schizophrenia. Psychiatry Res 2018;265:25–38.

54. Sliwoski G, Schubert M, Stichel J, Weaver D, Beck-Sickinger AG, Meiler J. Discovery of small-molecule modulators of the human Y4 receptor. PLoS One 2016;11:e0157146.

55. DE Hert M, Schreurs V, Vancampfort D, VAN Winkel R. Metabolic syndrome in people with schizophrenia: a review. World Psychiatry 2009;8:15–22.

56. Stuart MJ, Baune BT. Chemokines and chemokine receptors in mood disorders, schizophrenia, and cognitive impairment: a systematic review of biomarker studies. Neurosci Biobehav Rev 2014;42:93–115.

57. Maeda K, Das D, Nakata H, Mitsuya H. CCR5 inhibitors: emergence, success, and challenges. Expert Opin Emerg Drugs 2012;17:135–145.

58. Kalkonde YV, Shelton R, Villarreal M, Sigala J, Mishra PK, Ahuja SS, et al. The CC chemokine receptor 5 regulates olfactory and social recognition in mice. Neuroscience 2011;197:153–161.

59. Dasdemir S, Kucukali CI, Bireller ES, Tuzun E, Cakmakoglu B. Chemokine gene variants in schizophrenia. Nord J Psychiatry 2016;70:407–412.

60. Rasmussen HB, Timm S, Wang AG, Søeby K, Lublin H, Fenger M, et al. Association between the CCR5 32-bp deletion allele and late onset of schizophrenia. Am J Psychiatry 2006;163:507–511.

61. Di Nuzzo S, Zanni M, De Panfilis G. Exacerbation of paranoid schizophrenia in a psoriatic patient after treatment with cyclosporine A, but not with etanercept. J Drugs Dermatol 2007;6:1046–1047.

62. Hauben M. Cyclosporine neurotoxicity. Pharmacotherapy 1996;16:576–583.

63. Sato Y, Takayanagi Y, Onaka T, Kobayashi E. Impact of cyclosporine upon emotional and social behavior in mice. Transplantation 2007;83:1365–1370.

64. Wijdicks EF, Wiesner RH, Krom RA. Neurotoxicity in liver transplant recipients with cyclosporine immunosuppression. Neurology 1995;45:1962–1964.

65. St Clair D, Blackwood D, Muir W, Walker M, Carothers A, Spowart G, et al. Association within a family of a balanced autosomal translocation with major mental illness. Lancet 1990;336:13–16.

66. Kim Y, Lee JS, Joo YH. Rapamycin increases the incidence of neuropsychiatric illness in kidney transplant patients through the suppression of neural stem cells. Transl Psychiatry 2020;10:156

67. Şenay O, Balcioglu YH, Ilnem MC. Possible involvement of inhibited mTOR signaling in de novo psychosis following the initiation of everolimus in a patient with tuberous sclerosis: a clinical vignette. Psychiatr Clin Psychopharmacol 2019;29:262–263.

68. Yui K, Imataka G, Sasaki H, Kawasaki Y, Okanshi T, Shiroki R, et al. Improvement in impaired social cognition but not seizures by everolimus in a child with tuberous sclerosis-associated autism through increased serum antioxidant proteins and oxidant/antioxidant status. Case Rep Pediatr 2019;2019:2070619

69. Kim Y, Noh YW, Kim K, Kim E. Hyperactive ACC-MDT pathway suppresses prepulse inhibition in mice. Schizophr Bull 2021;47:31–43.

70. Kim Y, Yang E, Kim H. Impaired prepulse inhibition in mice with IRSp53 deletion in modulatory neurotransmitter neurons including dopamine, acetylcholine, oxytocin, and serotonin. Biochem Biophys Res Commun 2022;586:114–120.

71. Spark DL, Fornito A, Langmead CJ, Stewart GD. Beyond antipsychotics: a twenty-first century update for preclinical development of schizophrenia therapeutics. Transl Psychiatry 2022;12:147

72. Orlovska-Waast S, Köhler-Forsberg O, Brix SW, Nordentoft M, Kondziella D, Krogh J, et al. Cerebrospinal fluid markers of inflammation and infections in schizophrenia and affective disorders: a systematic review and meta-analysis. Mol Psychiatry 2019;24:869–887.

73. Upthegrove R, Khandaker GM. Cytokines, Oxidative Stress and Cellular Markers of Inflammation in Schizophrenia. In: Khandaker G, Meyer U, Jones P, editor. Neuroinflammation and Schizophrenia. Current Topics in Behavioral Neurosciences, vol 44. Cham: Springer, 2019, p. 49–66.