Reviewed 17 May 2026

Example research study

Turning LLPS Biology into Drug Targets: AI-Generated Evidence Synthesis for Biomolecular Condensates

Biomolecular condensate modulators target proteins like β-catenin, MYC, and TDP-43 that lack conventional binding pockets — but the literature now spans polymer physics, structural biology, neurology, and oncology. We used BioSkepsis to synthesise 42 PubMed-verified studies into a complete grant-writing workflow: landscape review, mechanistic links, testable hypothesis, and lab-ready protocol — with every citation triple-verified.

TL;DR Biomolecular condensate drug discovery is one of the fastest-moving fields in pharmacology, with clinical candidates like Omomyc (MYC, Phase I/II), DNL343 (TDP-43/ISR, Phase 1b), and Quinacrine (TopBP1, preclinical) advancing against previously undruggable targets. BioSkepsis processed a single research query into five grant-ready outputs — a verified literature review, a research landscape synthesis, a 21-row mechanistic links table, a falsifiable hypothesis with study design, and a detailed experimental methodology — each with three-stage citation verification that flagged 15 unverified references and suggested alternatives. The result: a grant proposal backbone that would take weeks to assemble manually, delivered in minutes with a verified evidence trail.

Why biomolecular condensate drug targets overwhelm traditional literature review

Biomolecular condensates are membrane-less assemblies that concentrate proteins and nucleic acids through liquid-liquid phase separation (LLPS). They regulate gene expression, signal transduction, and stress responses — and their dysregulation drives cancer, ALS, and frontotemporal dementia (PMID: 28225081).

The therapeutic premise is compelling: proteins like MYC, β-catenin, and TDP-43 lack enzymatic active sites or well-defined binding pockets, making them resistant to conventional small-molecule targeting (PMID: 34942444, 33812316). Condensate-modifying therapeutics bypass this limitation by exploiting the biophysics of phase separation itself — dissolving pathological aggregates, inducing beneficial condensation, or reprogramming condensate morphology (PMID: 39757214).

The problem for grant writers is scope. A single research question about condensate modulators touches the molecular grammar of FUS-family prion-like domains (PMID: 29961577), β-catenin phase separation in Wnt-driven cancers (PMID: 40593772), TDP-43 aggregation in ALS (PMID: 40825784), and DNA-damage checkpoint condensates in colorectal cancer (PMID: 41190242). No PI manually tracks 42 papers spanning these four sub-fields while also writing the proposal.

What BioSkepsis produced from a single research query on LLPS drug targets

The research thread started with one prompt: a request for a detailed review of condensate modulators as drug targets, covering LLPS biophysics, undruggable targets (β-catenin, MYC, TDP-43), clinical trial updates, mechanisms of action, and the biotech landscape. BioSkepsis returned five distinct outputs, each building on the previous one.

BioSkepsis outputs from a single condensate research query

Output	Grant section it maps to	Key metrics
Verified literature review	State of the art	12 PMIDs verified; 2 flagged as unverified with alternatives
Research landscape synthesis	Background & significance	42 studies; 3 evidence phases (2006–2025); 5 unverified citations flagged
Mechanistic links table	Preliminary data rationale	21 molecular interactions with effect direction, context, and PMID
Testable hypothesis	Specific aims	3 predictions; 2 falsification criteria; 3 unverified citations flagged
Experimental methodology	Research plan	7 protocol sections; Go/No-Go criteria; 8 unverified citations flagged

Each output underwent BioSkepsis’s three-stage citation verification. Across all five outputs, the system flagged 15 citations that failed entity matching, conclusion alignment, or mechanism validation — and suggested topic-matched alternatives for each. A general-purpose LLM would have presented all 15 as verified without distinction.

Three-stage citation verification: how BioSkepsis catches what LLMs miss in LLPS reviews

The verification pipeline runs three independent checks on every PMID-claim pair. Consider a claim from the landscape synthesis: that TDP-43 and G3BP1 serve as primary hubs linking early structural studies to recent drug screening platforms.

This is not an edge case. Across the five outputs, BioSkepsis identified mismatches including: a paper on Argonaute specificity cited for P-granule behaviour (entity failure); a β-catenin condensate study cited for TDP-43 biomarker diagnostics (conclusion failure); and a G3BP1 covalent modification paper cited for cGAMP measurement it never performed (mechanism failure). Each flagged citation included a percentage-scored alternative from the same corpus.

Clinical-stage condensate therapeutics: what the verified evidence actually shows

The research thread identified three clinical-stage candidates with verified evidence trails — and one notable gap.

Omomyc (OMO-103) is a mini-protein that competes with c-MYC for DNA binding, representing the first direct MYC inhibitor to reach Phase I/II clinical trials in patients with advanced solid tumours including colorectal and triple-negative breast cancer (PMID: 34942444).

DNL343 is a small-molecule eIF2B activator designed to inhibit the integrated stress response downstream of TDP-43 pathology. Phase 1b data in ALS participants showed approximately 80% median reduction in ISR biomarkers (ATF4, CHAC1) in PBMCs and reduced CSF GDF-15 levels, though the HEALEY ALS Platform Trial did not meet primary clinical endpoints after six months (PMID: 40825784).

Quinacrine, an FDA-approved drug, was identified through high-throughput optogenetic screening as an inhibitor of TopBP1 condensates. In a mouse model of peritoneal carcinomatosis, it enhanced the efficacy of 5-fluorouracil and irinotecan by dampening the S-phase checkpoint (PMID: 41190242).

DPTX3186, initially requested in the research query, returned no supporting evidence in the 42-paper corpus. BioSkepsis flagged this explicitly as “NR” (not reported) rather than fabricating a plausible narrative — a critical distinction for grant writers who need to know where the evidence stops.

From PubMed corpus to mechanistic links: 21 molecular interactions mapped for grant rationale

Grant reviewers want to see that your specific aims emerge from established mechanistic evidence. BioSkepsis extracted 21 molecular factor–target pairs from the corpus, each annotated with link type (phosphorylation, covalent modification, LLPS induction), effect direction (activating, inhibitory, regulatory), biological context, and source PMID.

Selected mechanistic links from the BioSkepsis-generated table

Molecular factor	Target	Effect	Mechanism	PMID
Rosmanol quinone	β-catenin	Inhibitory	Induces cytoplasmic LLPS to block nuclear translocation	40593772
CGX-635	Fibrillarin	Inhibitory	Converts liquid condensates to solid aggregates; reduces MYC translation	38924716
CAA-AA	G3BP1 Cys73	Inhibitory	Covalent modification promotes closed conformation; blocks stress granule formation	40751677
DNL343	eIF2B	Regulatory	Activates eIF2B to modulate ISR and reduce ATF4 in ALS models	40825784
Quinacrine	TopBP1	Inhibitory	Blocks TopBP1 chromatin association; inhibits condensate-mediated ATR/Chk1 signalling	41190242
ANXA11 LCD	TDP-43 LCD	Structural	Drives heteromeric amyloid filament co-assembly in FTLD-TDP type C	39260416

The complete 21-row table is available in the source research thread. Each row functions as a line of preliminary evidence in a grant application — with the PMID already verified against the specific claim it supports.

Hypothesis generation: from verified LLPS mechanisms to falsifiable predictions

BioSkepsis synthesised the mechanistic links into a specific, testable hypothesis: that covalent modification of G3BP1 at Cys73 using carbonylacrylic amide derivatives (CAA-AA) would selectively disrupt TDP-43 recruitment to stress granules, prevent the liquid-to-solid transition driving neurotoxicity, and ameliorate cGAS/STING-mediated neuroinflammation.

The hypothesis includes three explicit predictions (persistent TDP-43 puncta reduction in TARDBP-mutant iPSC-derived motor neurons; decreased cytosolic cGAMP; rescued neurite outgrowth), two falsification criteria (no effect in C73A-G3BP1 mutant neurons would indicate off-target activity; failure to prevent mtDNA leakage despite puncta reduction would break the mechanistic chain), and a complete study design with iPSC differentiation, puromycin stress induction, and 24-hour recovery imaging.

Critically, BioSkepsis flagged three citations within the hypothesis as unverified. One claim about cGAMP and IFN-β levels cited a paper (PMID: 33031745) that uses different inhibitors (RU.521, H-151) than the CAA-AA derivatives proposed. The system noted this entity mismatch and suggested alternatives. A general-purpose LLM would have presented this as a seamless, fully supported logical chain.

Lab-ready protocol: experimental methodology with Go/No-Go criteria for condensate drug screening

The final output was a detailed experimental methodology covering model selection (iPSC-derived motor neurons with TARDBP mutations), sample size and power analysis (6 biological replicates, α = 0.05, 80% power), interventions (30–40 μM CAA-AA with puromycin stress), controls (non-reactive 2H_CAA, C73A-G3BP1 CRISPR lines, cycloheximide positive control), endpoints with Go/No-Go thresholds (≥30% reduction in persistent TDP-43 puncta; viability >80% by MTT assay), and statistical analysis (two-way ANOVA, Tukey’s post-hoc, Benjamini-Hochberg correction).

BioSkepsis flagged 8 citations within the methodology as unverified — primarily because the specific combination of CAA-AA with iPSC motor neuron models has not been published (the CAA-AA paper used HeLa cells; the iPSC protocols come from separate ALS studies). This transparent flagging tells the grant writer exactly where the proposed work extends beyond existing evidence — which is precisely what a novel research proposal should do.

Knowledge gaps in condensate drug discovery: label-free measurement and clinical translation

BioSkepsis identified four knowledge gaps from the corpus that represent opportunities for novel grant proposals. First, the failure of DNL343 to meet primary clinical endpoints despite strong biomarker engagement suggests a need for patient stratification or longer treatment duration in ISR-targeted ALS therapies (PMID: 40825784).

Second, label-free quantitative phase imaging (QPI) reveals that common fluorescent tags like mEGFP alter dense-phase concentrations by up to 14% and underestimate partition coefficients — meaning some earlier drug-condensate interaction data may need recalibration (PMID: 40903498).

Third, the discovery of heteromeric ANXA11-TDP-43 amyloid filaments in FTLD-TDP type C (PMID: 39260416) raises questions about which cellular stressors trigger the liquid-to-solid transition in vivo — a mechanism that remains poorly defined.

Fourth, fluid biomarkers like cryptic HDGFL2 can detect TDP-43 dysfunction in presymptomatic ALS-FTD carriers (PMID: 38278991), opening a path toward earlier diagnosis and more precise clinical trial enrolment — but integration with condensate-targeting therapeutics has not been explored.

Who benefits: grant writers, PIs, and drug discovery teams in LLPS therapeutics

Frequently asked questions

What are biomolecular condensate modulators and why are they hard to review?

Condensate modulators (c-mods) are small molecules that dissolve, induce, or reprogram membrane-less protein assemblies formed by liquid-liquid phase separation. The field spans polymer physics, cell biology, and pharmacology, drawing on hundreds of studies across ALS, oncology, and structural biology — making manual literature synthesis extremely time-consuming.

How does BioSkepsis verify citations in condensate drug discovery reviews?

Every PMID undergoes three independent verification checks: entity matching (do the genes, proteins, and drugs in the claim appear in the paper?), conclusion alignment (does the paper’s conclusion support the specific claim?), and mechanism validation (does the described mechanism match?). Citations that fail any check are flagged as unverified with suggested alternatives.

Can BioSkepsis generate testable hypotheses for LLPS-targeted therapeutics?

Yes. BioSkepsis synthesises mechanistic links across verified studies and proposes empirically testable hypotheses with explicit predictions, study designs, statistical frameworks, and falsification criteria — all grounded in real PMIDs.

What grant sections can BioSkepsis help draft for condensate research proposals?

BioSkepsis outputs map directly onto standard grant sections: the landscape synthesis becomes your state-of-the-art; the mechanistic links table becomes your preliminary data rationale; the hypothesis becomes your specific aims; and the methodology becomes your research plan — each with citation-grounded evidence trails.

How does BioSkepsis compare to ChatGPT or Gemini for biomedical literature reviews?

General-purpose LLMs generate plausible-sounding reviews but cannot verify whether a PMID actually supports a specific claim. BioSkepsis runs three-stage citation verification, flags unverified references, and suggests topic-matched alternatives — producing reviews where every claim traces to a confirmed source.

What clinical-stage condensate drug candidates does the research thread cover?

The thread covers Omomyc (OMO-103), a direct MYC inhibitor in Phase I/II for advanced solid tumours (PMID: 34942444); DNL343, an eIF2B activator tested in Phase 1 and Phase 1b ALS trials (PMID: 40825784); and Quinacrine, an FDA-approved drug repurposed as a TopBP1 condensate inhibitor that enhances FOLFIRI efficacy in colorectal cancer models (PMID: 41190242).

Is BioSkepsis free to use for grant writing in life sciences?

BioSkepsis offers a free tier that includes research threads with citation verification. Researchers can start a thread, receive a verified literature synthesis, and use the outputs directly in grant applications.

Run your own PubMed-verified condensate literature review

Start a BioSkepsis research thread on any LLPS target, drug candidate, or grant question. Every citation triple-verified; every claim traceable to a specific PMID.

Start free

Sources & further reading

1. Banani SF, Lee HO, Hyman AA, Rosen MK. Biomolecular condensates: organizers of cellular biochemistry. Nat Rev Mol Cell Biol. 2017;18(5):285-298. PMID: 28225081
2. Wang J, Choi J-M, Holehouse AS, et al. A Molecular Grammar Governing the Driving Forces for Phase Separation of Prion-like RNA Binding Proteins. Cell. 2018;174(3):688-699.e16. PMID: 29961577
3. Biesaga M, Frigolé-Vivas M, Salvatella X. Intrinsically disordered proteins and biomolecular condensates as drug targets. Curr Opin Chem Biol. 2021;62:90-100. PMID: 33812316
4. Llombart V, Mansour MR. Therapeutic targeting of “undruggable” MYC. EBioMedicine. 2021;75:103756. PMID: 34942444
5. Irwin KE, Jasin P, Braunstein KE, et al. A fluid biomarker reveals loss of TDP-43 splicing repression in presymptomatic ALS-FTD. Nat Med. 2024;30(2):382-393. PMID: 38278991
6. Jeon S, Jeon Y, Lim J-Y, et al. Emerging regulatory mechanisms and functions of biomolecular condensates: implications for therapeutic targets. Signal Transduct Target Ther. 2025;10(1):4. PMID: 39757214
7. Yan J, Liu H, Yang W, et al. Small-molecule-induced liquid-liquid phase separation suppresses the carcinogenesis of β-catenin. Nat Commun. 2025;16(1):5997. PMID: 40593772
8. Flores BN, Yu SB, Cohen IV, et al. Investigational eIF2B activator DNL343 modulates the integrated stress response in preclinical models of TDP-43 pathology and individuals with ALS. Nat Commun. 2025;16(1):7690. PMID: 40825784
9. McCall PM, Kim K, Shevchenko A, et al. A label-free method for measuring the composition of multicomponent biomolecular condensates. Nat Chem. 2025;17(12):1891-1902. PMID: 40903498
10. Morano L, Vie N, Aissanou A, et al. Shining light on drug discovery: optogenetic screening for TopBP1 biomolecular condensate inhibitors. NAR Cancer. 2025;7(4):zcaf041. PMID: 41190242