Decoding the dark proteome: Deep learning-enabled discovery of druggable enzymes in Wuchereria bancrofti

Shawnak Shivakumar, Jefferson HernandezarXiv 2025 · arXiv · OpenReview

TL;DR

This paper tackles the challenge of identifying drug targets in the neglected tropical disease parasite Wuchereria bancrofti by using deep learning to annotate its "dark proteome" – the vast majority of its proteins with unknown functions. They developed a novel computational pipeline that successfully identified 14,772 previously uncharacterized proteins with precise enzymatic functions and prioritized six promising drug targets with potential small-molecule inhibitors.

What Problem Does This Paper Address?

Wuchereria bancrofti is the primary cause of lymphatic filariasis, a debilitating disease affecting millions globally. A major hurdle in developing new treatments is the lack of functional understanding for over 90% of its proteins, often referred to as the "dark proteome." Without knowing what these proteins do, it's incredibly difficult to identify potential drug targets. Existing methods for protein function prediction often struggle with novel or poorly studied organisms, leaving a critical gap in our ability to discover new antiparasitic drugs.

Methods

The researchers developed a multi-stage computational pipeline:

  1. Enzyme vs. Non-Enzyme Classification: They employed a modified Protein Detection Transformer (ProtDETR). This model, originally designed for image analysis, was adapted to analyze amino acid sequences. It identifies functional segments within proteins and outputs a probability of enzymatic function, allowing them to filter out non-enzymes with high confidence (>98%).
  2. Four-Digit EC Number Assignment: For the identified enzymes, they used a Hierarchically-Finetuned Nearest Neighbor (HiFi-NN) model. This deep learning architecture combines transformer embeddings with a nearest-neighbor classifier to predict precise four-level Enzyme Commission (EC) numbers, which classify enzymes based on their biochemical reactions. They fine-tuned this model on existing W. bancrofti enzyme data.
  3. Target Prioritization: A multi-criteria approach was used to select promising drug targets from the newly annotated enzymes. This involved:
  4. Parasite-specific target: Ensuring minimal similarity to human homologs to reduce potential toxicity.
  5. Chemical tractability: Checking for existing compounds or inhibitors in databases like ChEMBL and BindingDB.
  6. Biochemical importance: Prioritizing enzymes involved in essential parasite processes.
  7. Biological plausibility: Confirming the enzyme's presence in related organisms.
  8. Binding Affinity Prediction: For the prioritized targets, they used Boltz-2, an AI-based molecular docking (co-folding) model, to predict binding affinities of small molecules. This allows for rapid screening of potential drug candidates.

Datasets

  • UniProt Database: The primary source for W. bancrofti protein sequences (Taxonomy ID: 6293).
  • Labeled Parasite Proteins: 4,476 W. bancrofti proteins with known EC annotations were used for training and fine-tuning the EC predictor.
  • ChEMBL and BindingDB: These databases were queried to identify existing compounds and inhibitors for enzyme classes.

Key Insights and Ideas

  • Deep Learning for "Dark Proteomes": The study demonstrates the power of deep learning, particularly transformer-based models, to unlock the functional potential of previously uncharacterized proteins in organisms with limited prior annotation.
  • EC Numbers as a Rosetta Stone: Precisely assigning four-digit EC numbers provides a standardized and informative way to understand enzyme function, going beyond vague annotations and directly revealing biochemical activities.
  • Prioritization is Key: Simply annotating is not enough; a robust, multi-criteria prioritization strategy is crucial for translating functional predictions into actionable drug discovery leads.
  • Symbiotic Relationships Revealed: The identification of bacteria-like enzymes highlights the intricate symbiotic relationship between W. bancrofti and its endosymbiont Wolbachia, offering potential avenues for targeting this partnership.

Key Results

  • Proteome-Wide Annotation: The pipeline successfully assigned precise four-digit EC numbers to 14,772 previously uncharacterized proteins.
  • Novel Enzyme Classes: 543 new EC classes were identified for W. bancrofti, with 73 of these having at least 10 distinct UniProt sequences in the parasite.
  • Enzyme-Rich Proteome: Approximately 85.8% of the W. bancrofti proteome was predicted to be enzymatic with high confidence.
  • Prioritized Drug Targets: Six key enzymes were prioritized across five strategic categories:
  • Anti-Wolbachia: Peptidoglycan glycosyltransferase (EC 2.4.99.28) with nanomolar inhibitors (e.g., Moenomycin A12).
  • Core Function: Cathepsin B (EC 3.4.22.1) and Peptidase 1 (EC 3.4.22.65) with sub-micromolar inhibitors (e.g., K777).
  • Life Cycle/Transmission: Chitinase (EC 3.2.1.14) with moderate predictions (e.g., Allosamidin).
  • Immune-Evasion: Nucleoside-triphosphate phosphatase (NTPase) (EC 3.6.1.15) with low-micromolar inhibitors (e.g., Suramin).
  • Metabolic/Neuromuscular Destabilization: cGMP Phosphodiesterase (EC 3.1.4.35) with moderate predictions (e.g., Mirodenafil).
  • Binding Affinities: All six prioritized targets showed at least moderately strong predicted binding affinities (<1 µM), with some candidates achieving nanomolar potency.

Limitations and Future Work

  • Experimental Validation Needed: The authors emphasize that experimental validation is crucial to confirm these findings and prioritize targets for further development.
  • Generalizability: While the pipeline is broadly applicable, W. bancrofti presented unique challenges due to a lack of prior data. Applying it to other organisms with similar data scarcity is a promising future direction.
  • Ligand Optimization: For some targets, the predicted binding affinities are moderate, suggesting that further ligand growth and optimization will be necessary to achieve robust sub-micromolar activity and parasite-specific profiles.
  • Generative Design Integration: The authors explored using generative design tools like DeepFrag for ligand optimization, noting that while promising, it requires careful human oversight to maintain pose and avoid structural clashes.

Summary updated

Abstract

Wuchereria bancrofti, the parasitic roundworm responsible for lymphatic filariasis, permanently disables over 36 million people and places 657 million at risk across 39 countries. A major bottleneck for drug discovery is the lack of functional annotation for more than 90 percent of the W. bancrofti dark proteome, leaving many potential targets unidentified. In this work, we present a novel computational pipeline that converts W. bancrofti's unannotated amino acid sequence data into precise four-level Enzyme Commission (EC) numbers and drug candidates. We utilized a DEtection TRansformer to estimate the probability of enzymatic function, fine-tuned a hierarchical nearest neighbor EC predictor on 4,476 labeled parasite proteins, and applied rejection sampling to retain only four-level EC classifications at 100 percent confidence. This pipeline assigned precise EC numbers to 14,772 previously uncharacterized proteins and discovered 543 EC classes not previously known in W. bancrofti. A qualitative triage emphasizing parasite-specific targets, chemical tractability, biochemical importance, and biological plausibility prioritized six enzymes across five separate strategies: anti-Wolbachia cell-wall inhibition, proteolysis blockade, transmission disruption, purinergic immune interference, and cGMP-signaling destabilization. We curated a 43-compound library from ChEMBL and BindingDB and co-folded across multiple protein conformers with Boltz-2. All six targets exhibited at least moderately strong predicted binding affinities below 1 micromolar, with moenomycin analogs against peptidoglycan glycosyltransferase and NTPase inhibitors showing promising nanomolar hits and well-defined binding pockets. While experimental validation remains essential, our results provide the first large-scale functional map of the W. bancrofti dark proteome and accelerate early-stage drug development for the species.
BibTeX
@misc{shivakumar2025decoding,
  author = {Shawnak Shivakumar and Jefferson Hernandez},
  title = {{Decoding the dark proteome: Deep learning-enabled discovery of druggable enzymes in Wuchereria bancrofti}},
  year = {2025},
  eprint = {2510.07337},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2510.07337},
}
Figure