Genome-wide DNA methylation-analysis of blastic plasmacytoid dendritic cell neoplasm identifies distinct molecular features.

Blastic plasmacytoid dendritic cell neoplasm (BPDCN) constitutes a rare and aggressive malignancy originating from plasmacytoid dendritic cells (pDCs) with a primarily cutaneous tropism followed by dissemination to the bone marrow and other organs. We conducted a genome-wide analysis of the tumor methylome in an extended cohort of 45 BPDCN patients supplemented by WES and RNA-seq as well as ATAC-seq on selected cases. We determined the BPDCN DNA methylation profile and observed a dramatic loss of DNA methylation during malignant transformation from early and mature DCs towards BPDCN. DNA methylation profiles further differentiate between BPDCN, AML, CMML, and T-ALL exhibiting the most striking global demethylation, mitotic stress, and merely localized DNA hypermethylation in BPDCN resulting in pronounced inactivation of tumor suppressor genes by comparison. DNA methylation-based analysis of the tumor microenvironment by MethylCIBERSORT yielded two, prognostically relevant clusters (IC1 and IC2) with specific cellular composition and mutational spectra. Further, the transcriptional subgroups of BPDCN (C1 and C2) differ by DNA methylation signatures in interleukin/inflammatory signaling genes but also by higher transcription factor activity of JAK-STAT and NFkB signaling in C2 in contrast to an EZH2 dependence in C1-BPDCN. Our integrative characterization of BPDCN offers novel molecular insights and potential diagnostic applications.

Late Onset of Primary Hemophagocytic Lymphohistiocytosis (HLH) with a Novel Constellation of Compound Heterozygosity Involving Two Missense Variants in the PRF1 Gene.

Hemophagocytic lymphohistiocytosis (HLH) is a rare but in most cases life-threatening immune-mediated disease of the hematopoietic system frequently associated with hematologic neoplasms. Here, we report on a case in which we detected a novel constellation of two missense variants affecting the PRF1 gene, leading to de novo primary HLH. Diagnostics included a comprehensive clinical work-up and standard methods of hematopathology as well as extended molecular genomics based on polymerase chain reaction (PCR) reactions and the calculation of three-dimensional molecule reconstructions of PRF1. Subsequently, a comprehensive review of the literature was performed, which showed that this compound heterozygosity has not been previously described. The patient was a 20-year-old female. Molecular diagnostics revealed two heterozygous missense variants in the PRF1 gene (A91V and R104C) on exon 2. Apart from the finding of two inconclusive genetic variants, all clinical criteria defined by the HLH study group of Histiocyte Society were met at initial presentation. The final diagnosis was made in cooperation with the Consortium of German HLH-reference centers. Here, chemotherapy did not lead to sufficient sustained disease control. Therefore, the decision for allogenic hematopoietic stem cell transplantation (alloHSCT) was made. Hitherto, the duration of response was 6 months. Due to severe and unmanageable hepatic graft-versus-host disease (GvHD), the patient died. We report on a novel constellation of a compound heterozygosity containing two missense variants on exon 2 of the PRF1 gene. To the authors' best knowledge, this is the first presentation of a primary HLH case harboring this genomic constellation with late-onset clinical manifestation.

Diagnostic management of blastic plasmacytoid dendritic cell neoplasm (BPDCN) in close interaction with therapeutic considerations.

Blastic plasmacytoid dendritic cell neoplasm (BPDCN), a rare malignancy derived from plasmacytoid dendritic cells, can mimic both acute leukemia and aggressive T-cell lymphoma. Therapy of this highly aggressive hematological disease should be initiated as soon as possible, especially in light of novel targeted therapies that have become available. However, differential diagnosis of BPDCN remains challenging. This retrospective study aimed to highlight the challenges to timely diagnoses of BPDCN. We documented the diagnostic and clinical features of 43 BPDCN patients diagnosed at five academic hospitals from 2001-2022. The frequency of BPDCN diagnosis compared to AML was 1:197 cases. The median interval from the first documented clinical manifestation to diagnosis of BPDCN was 3 months. Skin (65%) followed by bone marrow (51%) and blood (45%) involvement represented the most common sites. Immunophenotyping revealed CD4 + , CD45 + , CD56 + , CD123 + , HLA-DR + , and TCL-1 + as the most common surface markers. Overall, 86% (e.g. CD33) and 83% (e.g., CD7) showed co-expression of myeloid and T-cell markers, respectively. In the median, we detected five genomic alterations per case including mutational subtypes typically involved in AML: DNA methylation (70%), signal transduction (46%), splicing factors (38%), chromatin modification (32%), transcription factors (32%), and RAS pathway (30%), respectively. The contribution of patients (30%) proceeding to any form of upfront stem cell transplantation (SCT; autologous or allogeneic) was almost equal resulting in beneficial overall survival rates in those undergoing allogeneic SCT (p = 0.0001). BPDCN is a rare and challenging entity sharing various typical characteristics of other hematological diseases. Comprehensive diagnostics should be initiated timely to ensure appropriate treatment strategies.

Induction treatment in high-grade B-cell lymphoma with a concurrent MYC and BCL2 and/or BCL6 rearrangement: a systematic review and meta-analysis.

Background and aim: High-grade B cell lymphomas with concomitant MYC and BCL2 and/or BCL6 rearrangements (HGBCL-DH/TH) have a poor prognosis when treated with the standard R-CHOP-like chemoimmunotherapy protocol. Whether this can be improved using intensified regimens is still under debate. However, due to the rarity of HGBCL-DH/TH there are no prospective, randomized controlled trials (RCT) available. Thus, with this systematic review and meta-analysis we attempted to compare survival in HGBCL-DH/TH patients receiving intensified vs. R-CHOP(-like) regimens. Methods: The PubMed and Web of Science databases were searched for original studies reporting on first-line treatment in HGBCL-DH/TH patients from 08/2014 until 04/2022. Studies with only localized stage disease, ≤10 patients, single-arm, non-full peer-reviewed publications, and preclinical studies were excluded. The quality of literature and the risk of bias was assessed using the Methodological Index for Non-Randomized Studies (MINORS) and National Heart, Lung, and Blood Institute (NHLBI) Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies. Random-effect models were used to compare R-CHOP-(like) and intensified regimens regarding 2-year overall survival (2y-OS) and 2-year progression-free survival (2y-PFS). Results: Altogether, 11 retrospective studies, but no RCT, with 891 patients were included. Only four studies were of good quality based on aforementioned criteria. Intensified treatment could improve 2y-OS (hazard ratio [HR]=0.78 [95% confidence interval [CI] 0.63-0.96]; p=0.02) as well as 2y-PFS (HR=0.66 [95% CI 0.44-0.99]; p=0.045). Conclusions: This meta-analysis indicates that intensified regimens could possibly improve 2y-OS and 2y-PFS in HGBCL-DH/TH patients. However, the significance of these results is mainly limited by data quality, data robustness, and its retrospective nature. There is still a need for innovative controlled clinical trials in this difficult to treat patient population. Systematic review registration: https://www.crd.york.ac.uk/prospero, identifier CRD42022313234.

Molecularly Stratified Treatment Options in Primary Refractory DLBCL/HGBL with MYC and BCL2 or BCL6 Rearrangements (HGBL, NOS with MYC/BCL6).

BACKGROUND: There is growing evidence supporting multidisciplinary molecular tumor boards (MTB) in solid tumors whereas hematologic malignancies remain underrepresented in this regard. OBJECTIVE: The present study aimed to assess the clinical relevance of MTBs in primary refractory diffuse large B-cell lymphomas/high-grade B-cell lymphomas with MYC and BCL2 rearrangements (prDLBCL/HGBL-MYC/BCL2) (n = 13) and HGBL, not otherwise specified (NOS), with MYC and BCL6 rearrangements (prHGBL, NOS-MYC/BCL6) (n = 6) based on our previously published whole-exome sequencing (WES) cohort. PATIENTS AND METHODS: For genomic analysis, the institutional MTB WES pipeline (University Cancer Center Schleswig-Holstein: UCCSH), certified for routine clinical diagnostics, was employed and supplemented by a comprehensive immunohistochemical work-up. Consecutive database research and annotation according to established evidence levels for molecularly stratified therapies was performed (NCT-DKTK/ESCAT). RESULTS: Molecularly tailored treatment options with NCT-DKTK evidence level of at least m2A were identified in each case. We classified mutations in accordance with biomarker/treatment baskets and detected a heterogeneous spectrum of targetable alterations affecting immune evasion (IE; n = 30), B-cell targets (BCT; n = 26), DNA damage repair (DDR; n = 20), tyrosine kinases (TK; n = 13), cell cycle (CC; n = 7), PI3K-MTOR-AKT pathway (PAM; n = 2), RAF-MEK-ERK cascade (RME; n = 1), and others (OTH; n = 11). CONCLUSION: Our virtual MTB approach identified potential molecularly targeted treatment options alongside targetable genomic signatures for both prDLBCL/HGBL-MYC/BCL2 and prHGBL, NOS-MYC/BCL6. These results underline the potential of MTB consultations in difficult-to-treat lymphomas early in the treatment sequence.

SchNetPack 2.0: A neural network toolbox for atomistic machine learning.

SchNetPack is a versatile neural network toolbox that addresses both the requirements of method development and the application of atomistic machine learning. Version 2.0 comes with an improved data pipeline, modules for equivariant neural networks, and a PyTorch implementation of molecular dynamics. An optional integration with PyTorch Lightning and the Hydra configuration framework powers a flexible command-line interface. This makes SchNetPack 2.0 easily extendable with a custom code and ready for complex training tasks, such as the generation of 3D molecular structures.

Primary refractory plasmablastic lymphoma: A precision oncology approach.

Introduction: Hematologic malignancies are currently underrepresented in multidisciplinary molecular-tumor-boards (MTB). This study assesses the potential of precision-oncology in primary-refractory plasmablastic-lymphoma (prPBL), a highly lethal blood cancer. Methods: We evaluated clinicopathological and molecular-genetic data of 14 clinically annotated prPBL-patients from initial diagnosis. For this proof-of-concept study, we employed our certified institutional MTB-pipeline (University-Cancer-Center-Schleswig-Holstein, UCCSH) to annotate a comprehensive dataset within the scope of a virtual MTB-setting, ultimately recommending molecularly stratified therapies. Evidence-levels for MTB-recommendations were defined in accordance with the NCT/DKTK and ESCAT criteria. Results: Median age in the cohort was 76.5 years (range 56-91), 78.6% of patients were male, 50% were HIV-positive and clinical outcome was dismal. Comprehensive genomic/transcriptomic analysis revealed potential recommendations of a molecularly stratified treatment option with evidence-levels according to NCT/DKTK of at least m2B/ESCAT of at least IIIA were detected for all 14 prPBL-cases. In addition, immunohistochemical-assessment (CD19/CD30/CD38/CD79B) revealed targeted treatment-recommendations in all 14 cases. Genetic alterations were classified by treatment-baskets proposed by Horak et al. Hereby, we identified tyrosine-kinases (TK; n=4), PI3K-MTOR-AKT-pathway (PAM; n=3), cell-cycle-alterations (CC; n=2), RAF-MEK-ERK-cascade (RME; n=2), immune-evasion (IE; n=2), B-cell-targets (BCT; n=25) and others (OTH; n=4) for targeted treatment-recommendations. The minimum requirement for consideration of a drug within the scope of the study was FDA-fast-track development. Discussion: The presented proof-of-concept study demonstrates the clinical potential of precision-oncology, even in prPBL-patients. Due to the aggressive course of the disease, there is an urgent medical-need for personalized treatment approaches, and this population should be considered for MTB inclusion at the earliest time.

The expression of the adenosine pathway markers CD39 and CD73 in salivary gland carcinomas harbors the potential for novel immune checkpoint inhibition.

BACKGROUND: In salivary gland carcinomas (SGC), there is only a small fraction of entities that appears to profit from immune checkpoint inhibition (ICI). Recent findings connected the activation of adenosine-signaling with a tolerogenic microenvironment. Therefore, the inhibition of adenosine pathway markers (CD39 and/or CD73) can augment ICI and/or display a novel immunotherapeutic strategy beyond ICI. Here, we assessed the immuno-histochemical expression of CD39 and CD73 across a wide spectrum of SGCs. METHODS: In total, 114 patients with SGCs consecutively diagnosed between 2001 and 2021 were assessed for clinicopathological baseline characteristics and underwent confirmatory histopathological review. Immunohistochemical expression levels of CD39 and CD73 were assessed by applying the tumor proportion score (TPS) and the immune proportional score (IPS) comparable to PD-L1 expression analysis in routine clinical practice. Additionally, findings were correlated with PD-L1 expression levels. RESULTS: The median age was 60.6 and 51.8% patients were female. The cohort covered a spectrum of eight distinct entities. Advanced-stage disease (UICC/AJCC III/IVA-IVC) at initial diagnosis was present in the majority of patients (64/114). Immunohistochemical staining revealed positivity for CD39 and CD73 in 48.2% and 21.1% on tumor cells (TPS ≥ 1%) as well as 46.4% and 42.9% within the immune cell infiltrate (IPS ≥ 1%), respectively. Further comparative analyses revealed immune-cold entities such adenoid cystic carcinoma (AdCC), immune-hot tumors such as adenocarcinoma, not otherwise specified (AC (NOS)) and entities with intermediate immunologic features such as acinic cell carcinoma (ACC). CONCLUSION: Current results indicate entity-specific adenosine signaling signatures. These findings suggest that the adenosine pathway plays a decisive role in tumor immunity among the major spectrum of SGCs. Targeting the adenosine pathway might pose a promising therapeutic option for selected entities.

Multifocal and hormone-dependent epithelioid hemangioendothelioma with osteolysis of the second cervical vertebral body: report of an unprecedented surgical approach by using autologous bone graft.

We report the case of a 28-year-old female patient who complained of extreme neck pain when giving birth to a child. Magnetic resonance imaging (MRI) of the cervical spine demonstrated an osteolytic lesion at the second cervical vertebral body (C2). In this presentation, we highlight a transoral surgical approach in order to prevent instability of this osteolytic lesion. To the best of our knowledge, this is the first time that such a route of access has been described for this tumor entity. A histopathologic examination led to the diagnosis of epithelioid hemangioendothelioma. During a follow-up period of 33 months, the patient had no complaints.

The Glasgow Prognostic Score Predicts Survival Outcomes in Neuroendocrine Neoplasms of the Gastro-Entero-Pancreatic (GEP-NEN) System.

Background: Across a variety of solid tumors, prognostic implications of nutritional and inflammation-based risk scores have been identified as a complementary resource of risk stratification. Methods: In this retrospective study, we performed a comparative analysis of several established risk scores and ratios, such as the Glasgow Prognostic Score (GPS), in neuroendocrine neoplasms of the gastro−entero−pancreatic (GEP-NEN) system with respect to their prognostic capabilities. Clinicopathological and treatment-related data for 102 GEP-NEN patients administered to the participating institutions between 2011 and 2021 were collected. Scores/ratios significantly associated with overall or progression-free survival (OS, PFS) upon univariate analysis were subsequently included in a Cox-proportional hazard model for the multivariate analysis. Results: The median age was 62 years (range 18−95 years) and the median follow-up period spanned 51 months. Pancreatic or intestinal localization at the initial diagnosis were present in 41 (40.2%) and 44 (43.1%) cases, respectively. In 17 patients (16.7%), the primary manifestation could not be ascertained (NNUP; neuroendocrine neoplasms of unknown primary). Histological grading (HG) revealed 24/102 (23.5%) NET/NEC (poorly differentiated; high grade G3) and 78/102 (76.5%) NET (highly or moderately differentiated; low−high grade G1−G2). In total, 53/102 (51.9%) patients presented with metastatic disease (UICC IV), 11/102 (10.7%) patients presented with multifocal disease, and 56/102 (54.9%) patients underwent a primary surgical or endoscopic approach, whereas 28 (27.5%) patients received systemic cytoreductive treatment. The univariate analysis revealed the GPS and PI (prognostic index), as well as UICC-stage IV, HG, and the Charlson comorbidity index (CCI) to predict both the PFS and OS in GEP-NEN patients. However, the calculation of the survival did not separate GPS subgroups at lower risk (GPS 0 versus GPS 1). Upon the subsequent multivariate analysis, GPS was the only independent predictor of both OS (p < 0.0001; HR = 3.459, 95% CI = 1.263−6.322) and PFS (p < 0.003; HR = 2.119, 95% CI = 0.944−4.265). Conclusion: In line with previous results for other entities, the present study revealed the GPS at baseline to be the only independent predictor of survival across all stages of GEP-NEN, and thus supports its clinical utility for risk stratification in this group of patients.

Inverse design of 3d molecular structures with conditional generative neural networks.

The rational design of molecules with desired properties is a long-standing challenge in chemistry. Generative neural networks have emerged as a powerful approach to sample novel molecules from a learned distribution. Here, we propose a conditional generative neural network for 3d molecular structures with specified chemical and structural properties. This approach is agnostic to chemical bonding and enables targeted sampling of novel molecules from conditional distributions, even in domains where reference calculations are sparse. We demonstrate the utility of our method for inverse design by generating molecules with specified motifs or composition, discovering particularly stable molecules, and jointly targeting multiple electronic properties beyond the training regime.

Mutational and immunologic Landscape in malignant Salivary Gland Tumors harbor the potential for novel therapeutic strategies.

Salivary gland carcinomas (SGC) are rare (3-6 % of all head and neck cancers) and show biological heterogeneity depending on the respective histological subtype. While complete surgical resection is the standard treatment for localized disease, chemotherapy or radiation therapy are frequently insufficient for the treatment of unresectable or metastasized SGC. Therefore, new therapeutic approaches such as molecularly targeted therapy or the application of immune checkpoint inhibition enhance the treatment repertoire. Accordingly, comprehensive analyses of the genomic landscape and the tumor-microenvironment (TME) are of crucial importance in order to optimize and individualize SGC treatment. This manuscript combines the current scientific knowledge of the composition of the mutational landscape and the TME in SGCs harboring the potential for novel (immune-) targeted therapeutic strategies.

Prognostic impact of nutritional and inflammation-based risk scores in follicular lymphoma in the era of anti-CD20 targeted treatment strategies.

BACKGROUND: The composition of the tumor microenvironment (TME) is conditioned by immunity and the inflammatory response. Nutritional and inflammation-based risk scores have emerged as relevant predictors of survival outcome across a variety of hematological malignancies. METHODS: In this retrospective multicenter trial, we ascertained the prognostic impact of established nutritional and inflammation-based risk scores [Glasgow Prognostic Score (GPS), C-reactive-protein/albumin ratio (CAR), neutrophil-lymphocyte ratio (NLR), prognostic nutritional index (PNI), and prognostic index (PI)] in 209 eligible patients with histologically confirmed CD20+ follicular lymphoma (FL) of WHO grade 1 (37.3%), 1-2 (16.3%), 2 (26.8%) or 3A (19.8%) admitted to the participating centers between January 2000 and December 2019. Characteristics significantly associated with overall or progression-free survival (OS, PFS) upon univariate analysis were subsequently included in a Cox proportional hazard model. RESULTS: In the study cohort, the median age was 63 (range 22-90 years). The median follow-up period covered 99 months. The GPS and the CAR were identified to predict survival in FL patients. The GPS was the only independent predictor of OS (p < 0.0001; HR 2.773; 95% CI 1.630-4.719) and PFS (p = 0.001; HR 1.995; 95% CI 1.352-2.944) upon multivariate analysis. Additionally, there was frequent occurrence of progression of disease within 24 months (POD24) in FL patients with a calculated GPS of 2. CONCLUSION: The current results indicate that the GPS predicts especially OS in FL patients. Moreover, GPS was found to display disease-specific effects in regard to FL progression. These findings and potential combinations with additional established prognosticators should be further validated within prospective clinical trials.

Mutational landscape of high-grade B-cell lymphoma with MYC-, BCL2 and/or BCL6 rearrangements characterized by whole-exome sequencing.

High-grade B-cell lymphoma accompanied with double/triple-hit MYC and BCL2 and/or BCL6 rearrangements (HGBLDH/ TH) poses a cytogenetically-defined provisional entity among aggressive B-cell lymphomas that is traditionally associated with unfavorable prognosis. In order to better understand the mutational and molecular landscape of HGBLDH/ TH we here performed whole-exome sequencing and deep panel next-generation sequencing of 47 clinically annotated cases. Oncogenic drivers, mutational signatures and perturbed pathways were compared with data from follicular lymphoma (FL), diffuse large B-cell lymphoma (DLBCL) and Burkitt lymphoma (BL). We find an accumulation of oncogenic mutations in NOTCH, IL6/JAK/STAT and NFκB signaling pathways and delineate the mutational relationship within the continuum between FL/DLBCL, HGBL-DH/TH and BL. Further, we provide evidence of a molecular divergence between BCL2 and BCL6 rearranged HGBL-DH. Beyond a significant congruency with the C3/EZB DLBCL cluster in BCL2 rearranged cases on an exome-wide level, we observe an enrichment of the SBS6 mutation signature in BCL6 rearranged cases. Differential gene set enrichment and subsequent network propagation analysis according to cytogenetically defined subgroups revealed an impairment of TP53 and MYC pathway signaling in BCL2 rearranged cases, whereas BCL6 rearranged cases lacked this enrichment, but instead showed impairment of E2F targets. Intriguingly, HGBL-TH displayed intermediate mutational features considering all three aspects. This study elucidates a recurrent pattern of mutational events driving FL into MYC-driven BCL2-rearranged HGBL, unveiling the mutational pathogenesis of this provisional entity. Through this refinement of the molecular taxonomy for aggressive, germinal center-derived B-cell lymphomas, this calls into question the current World Health Organization classification system, especially regarding the status of MYC/BCL6- rearranged HGBL.

Comparative analysis of international prognostic indices in gray-zone lymphoma.

Gray-zone lymphoma (GZL) reflects an aggressive B-cell neoplasm with features intermediate between diffuse large B-cell lymphoma (DLBCL) and classical Hodgkin lymphoma (cHL). The International Prognostic Index (IPI) and its derivatives (R-IPI, NCCN-IPI, and the Hasenclever IPS) have been established for DLBCL or cHL while the most suitable scoring system for GZL remains undetermined. In an exploratory multi-centric cohort of GZL (n = 61), we performed a comparative analysis of prognostic indices with regard to model fit and mutual concordance. The calculation of the corrected Akaike's information criterion (cAIC) and Harrel's concordance index (c-index) for each scoring system identified the NCCN-IPI to harbor the most convincing prognostic capabilities regarding both overall survival (OS) and progression-free survival (PFS) compared to its enhanced derivatives. The current results affirm the clinical utility of the NCCN-IPI and suggest its preferential use in clinical practice in GZL-patients.

Integrative genomic and transcriptomic analysis in plasmablastic lymphoma identifies disruption of key regulatory pathways.

Plasmablastic lymphoma (PBL) represents a clinically heterogeneous subtype of aggressive B-cell non-Hodgkin lymphoma. Targeted-sequencing studies and a single-center whole-exome sequencing (WES) study in HIV-positive patients recently revealed several genes associated with PBL pathogenesis; however, the global mutational landscape and transcriptional profile of PBL remain elusive. To inform on disease-associated mutational drivers, mutational patterns, and perturbed pathways in HIV-positive and HIV-negative PBL, we performed WES and transcriptome sequencing (RNA-sequencing) of 33 PBL tumors. Integrative analysis of somatic mutations and gene expression profiles was performed to acquire insights into the divergent genotype-phenotype correlation in Epstein-Barr virus-positive (EBV+) and EBV- PBL. We describe a significant accumulation of mutations in the JAK signal transducer and transcription activator (OSMR, STAT3, PIM1, and SOCS1), as well as receptor tyrosine-kinase RAS (ERBB3, NRAS, PDGFRB, and NTRK) pathways. We provide further evidence of frequent perturbances of NF-κB signaling (NFKB2 and BTK). Induced pathways, identified by RNA-sequencing, closely resemble the mutational profile regarding alterations accentuated in interleukin-6/JAK/STAT signaling, NF-κB activity, and MYC signaling. Moreover, class I major histocompatibility complex-mediated antigen processing and cell cycle regulation were significantly affected by EBV status. An almost exclusive upregulation of phosphatidylinositol 3-kinase/AKT/mTOR signaling in EBV+ PBL and a significantly induced expression of NTRK3 in concert with recurrent oncogenic mutations in EBV- PBL hint at a specific therapeutically targetable mechanism in PBL subgroups. Our characterization of a mutational and transcriptomic landscape in PBL, distinct from that of diffuse large B-cell lymphoma and multiple myeloma, substantiates the pathobiological independence of PBL in the spectrum of B-cell malignancies and thereby refines the taxonomy for aggressive lymphomas.

3D-Scaffold: A Deep Learning Framework to Generate 3D Coordinates of Drug-like Molecules with Desired Scaffolds.

The prerequisite of therapeutic drug design and discovery is to identify novel molecules and developing lead candidates with desired biophysical and biochemical properties. Deep generative models have demonstrated their ability to find such molecules by exploring a huge chemical space efficiently. An effective way to generate new molecules with desired target properties is by constraining the critical fucntional groups or the core scaffolds in the generation process. To this end, we developed a domain aware generative framework called 3D-Scaffold that takes 3D coordinates of the desired scaffold as an input and generates 3D coordinates of novel therapeutic candidates as an output while always preserving the desired scaffolds in generated structures. We demonstrated that our framework generates predominantly valid, unique, novel, and experimentally synthesizable molecules that have drug-like properties similar to the molecules in the training set. Using domain specific data sets, we generate covalent and noncovalent antiviral inhibitors targeting viral proteins. To measure the success of our framework in generating therapeutic candidates, generated structures were subjected to high throughput virtual screening via docking simulations, which shows favorable interaction against SARS-CoV-2 main protease (Mpro) and nonstructural protein endoribonuclease (NSP15) targets. Most importantly, our deep learning model performs well with relatively small 3D structural training data and quickly learns to generalize to new scaffolds, highlighting its potential application to other domains for generating target specific candidates.