PRMT5 in multiple cancers and AI-driven hit identification from 1.5M compounds

PRMT5 expression, alteration, and prognosis across tumor types

Investigators reported a pan-cancer characterization of PRMT5 using multiple public bioinformatics platforms. The study team queried TIMER3, GEPIA2, UALCAN/CPTAC, the Human Protein Atlas, cBioPortal, STRING and Enrichr to map mRNA expression, protein abundance, genetic alteration and co-expression networks. The analyses showed pervasive expression of PRMT5 across cancers but with tumor-specific patterns: the cohort showed tumor-type dependent prognostic impact, and the authors called out glioblastoma multiforme (GBM), lower grade glioma (LGG) and kidney renal clear cell carcinoma (KIRC) as contexts with notable associations between PRMT5 metrics and survival. Data indicated links between PRMT5 status and measures of immune infiltration and co-expression partners in some tumor types, though the preprint did not present biochemical or clinical endpoint data beyond these in silico associations.

Short sentence: this was a computational survey. Longer sentence: by cross-referencing expression, proteomic datasets and alteration profiles, the study team aimed to define where PRMT5 might exert context-dependent oncogenic or prognostic roles that could inform target selection for inhibitor development.

AI virtual screen: model, dataset size and prioritized candidates

The authors applied a deep neural network to virtually screen chemical space. They reported that about 1.5 million PubChem compounds were screened by the AI pipeline and that the deep neural network achieved a ROC-AUC = 0.9857 on the internal evaluation cited. From that large set, three candidate PubChem identifiers were prioritized: CID 44345182, CID 2794188 and CID 44345334. The study team described this screening as a high-throughput prioritization step feeding downstream structural evaluation.

The pipeline combined cheminformatics filtering with predictive scoring. Short statement: model discrimination was reported as high. Longer statement: the source did not report external prospective validation, biochemical inhibition constants, or cell-based activity for these candidates; those data remain absent in the preprint and will be required to move any nominee forward.

Molecular docking and comparative binding scores against PRMT5 (PDB ID: 4X61)

After AI prioritization, investigators docked the candidates against a PRMT5 crystal structure (PDB ID: 4X61). Docking results placed CID 44345182 as the top-scoring ligand with a reported docking score of -9.8 kcal/mol. The reference ligand used for comparison scored -9.2 kcal/mol in the same docking protocol. The study team observed that the candidate’s docking score was similar to or marginally better than the reference ligand under their reported parameters.

Longer sentence: the authors described docking poses and binding affinity estimates as part of an integrated selection cascade, while also emphasizing that docking scores alone do not confirm biochemical inhibition, a limitation the preprint explicitly acknowledged. Short sentence: docking was comparative, not definitive.

ADMET, toxicity filtering and 100 ns MD trajectories that informed hit selection

Selected compounds then underwent ADMET filtering and toxicity prediction steps, according to the preprint. The authors reported subsequent 100 ns molecular dynamics (MD) simulations for trajectory stability assessment. Data indicated that CID 44345182 showed trajectory stability and favorable behavior across the reported simulation window, supporting its nomination as the potential hit from this in silico pipeline.

The source stated investigators performed toxicity prediction but did not include exhaustive experimental toxicology or pharmacokinetic measurements. Longer sentence: while computational ADMET and MD can prioritize molecules and reveal potential liabilities or dynamic binding features, the study team noted that these outputs require biochemical and cellular confirmation before any translational claims can be made.

Study limitations, conflicts of interest and required experimental validation

The preprint explicitly states that further experimental validation is needed to establish CID 44345182 as a promising PRMT5 inhibitor. The source did not report enzymatic IC50 values, target engagement in cells, off-target screening, or in vivo efficacy data. Those endpoints remain unaddressed in the document and represent the principal next steps required to move a computational hit toward a lead compound.

The authors disclosed a competing interest: R.A.A.-A. is the founder of OxUpp and a founder and shareholder of PharmaPrecision. The remaining authors declared no financial interests or personal relationships that could have influenced the work, as noted in the preprint.

Short sentence: the work is hypothesis-generating. Longer sentence: the study provides a combined pan-cancer target characterization and an AI-driven prioritization workflow, yet it stops at in silico nomination and therefore cannot substitute for biochemical, cellular or animal model validation that would be necessary to confirm mechanism, potency, selectivity and safety.

Specific unanswered clinical and experimental questions listed in the source that must be addressed next include: biochemical inhibition potency (IC50), cellular activity and target engagement assays, off-target profiling, pharmacokinetics and in vivo efficacy and toxicity studies. The preprint did not report these data, so those remain concrete gaps for follow-up research.