iiinova — where your digital self keeps evolving

Radiant DIA: A Fast, Sensitive, and Accurate Search Engine for Quantitative Proteomics

Preprint: www.biorxiv.org/content/10.6... Repositories: github.com/seerbio GUI repository: github.com/seerbio/radi... Huge congratulations to everyone who helped build, test, document, and release this work. I’m excited to see how the community uses these tools to do fantastic science.

biorxiv.org

In mass spectrometry-based proteomics, robust and efficient search engines are essential for accurate peptide and protein identification and quantification. Advances in sample preparation and instrume...

Our preprint highlights the performance and engine design, but the software release is what makes this especially exciting to me: the pipeline is available at no cost under its license, the code can be inspected, and the project is open for feedback, bug reports, and community contributions.

Radiant DIA™ is our DIA search engine built for peptide-centric search. Fulcrum is the scalable workflow layer that helps run analyses from desktop-scale studies to larger cohort-scale processing, with flexible plugin capabilities for proteomics workflows.

Lee Cantrell @leecantrell.bsky.social · 03/06/2026

DIA proteomics needs more than powerful search — it needs transparent, reproducible, and scalable software. I’m excited to share that Radiant DIA™ and the Fulcrum Scalable Pipeline™ are now released with publicly available source code.

1142

Great preprint! Was cool to see the impact of Carafe and the implication that has towards predicted model optimization for DIA.

Lee Cantrell @leecantrell.bsky.social · 31/05/2026

The Talus team represents incredibly well. I’m always impressed.

Stay tuned for tweetorial on Radiant DIA :) Update coming later this week. (Spoiler: license terms & source availability that I think the community will appreciate)

020

8/ IDs are not the output to biology. Well-quantified ions are. Power can occur with reasonably balanced, appropriately sized cohorts at realistic precision, but ID expansion should rely on high-quality acquisition. Benchmark MS methods by LOQ yield, peak shape, and sample context, not IDs alone.

081

7/ In a 20 vs 20 plasma case-control cohort, reduced sampling preserved broad fold-change structure but eroded q-value concordance, achieved power, and BH-significant + powered precursors. Losses were enriched among lower-abundance features.

6/ Simulations recapitulated the pattern. Increasing DPPP mainly reduced integration variance. But integration accuracy remained governed by SNR, baseline/noise structure, and peak asymmetry. More scans help; they do not rescue poor chromatography or weak signal.

5/ DPPP was useful but incomplete. Low DPPP did not mean peaks were non-quantitative. Integration accuracy and precision were more governed by peak area than by DPPP alone. Peak context mattered more than a single sampling proxy.

4/ Key observation: LOD was comparatively resilient to reduced DPPP, while LOQ degraded strongly. Precursors could remain detectable yet lose the linearity and precision needed for quantitative interpretation.

3/ We applied deterministic MS2 subsampling to emulate reduced peak sampling while holding the sample constant. This let us evaluate the relationship among identification, LOD/LOQ, quantitative precision, and statistical power rather than treating IDs as the endpoint.

2/ DPPP is often used as a proxy for quantification quality. Reduced DPPP can arise from accelerated gradients, narrower/more numerous DIA windows, altered chromatography, longer effective cycle times, or other method changes that reduce MS2 observations across a peak.

Lindsay K Pino @lindsaykpino.com · 17/05/2026

www.biorxiv.org/content/10.6... 1/ Proteomics method development often optimizes for IDs. But for cohort biology, IDs only matter if they remain quantitative, precise, and powered. We ask how chromatographic sampling density propagates from peaks → LOQ → discovery. 🧵

1204

Reposted by Lee Cantrell

I'm sure @leecantrell.bsky.social will do a real tweetorial, but my favorite: "Reduced peak completeness minimally affected detectability (LOD) but substantially degraded quantifiability (LOQ), indicating that identification-centric readouts can remain stable while quantitative utility declines."

1104

Reposted by Lee Cantrell

bioRxiv Bioinfo @biorxiv-bioinfo.bsky.social · 17/05/2026

From Peaks to Power: Systematic Evaluation of Chromatographic Sampling Reveals Determinants of Quantification and Biological Discovery in DIA Proteomics www.biorxiv.org/content/10.64898/20…

042

Lee Cantrell @leecantrell.bsky.social · 14/05/2026

I think this is also capturing the verticals and non-cohesion of sites for Thermo. Examples could be sample prep, software, columns, data processing, targeted v discovery, etc. Bruker + Biognosys groups seems to have more unified messaging by merging brands, at cost to bulk.

000

Lee Cantrell @leecantrell.bsky.social · 08/05/2026

I wonder if they are basically adapting cPILOT…

320

The ceiling is not going to be reached, but cohort size is correlated. Also think goal of search. Fragpipe and spectrum centric searches will find PTMs easier. Peptide centric searches are much faster but have more narrow scope

In effect, your FDR has upper ceiling of n files * first pass threshold (e.g., 1%). Without decoy management in second pass MBR you inflate FDR significantly.

A massive limitation of some algs is lack of decoy reporting. This undermines confidence of match and especially in MBR based search, inflates FDR. These aren’t detected at high risk in n=4 studies. But for n=100-100000 there are profound consequences.

Lee Cantrell @leecantrell.bsky.social · 16/02/2026

Best software is super subjective. Higher count algorithms tend to be less conservative for FDR. Lower count tend to be overly conservative. The folks at MSAID have good thoughts on this as do MSFragger/Fragmatics.

100

These are also pretty conservative loading masses. Comparison of 4 ng to 4000 ng may understandably be quite different. I’d imagine AGC is being used frequently throughout the study.

Lee Cantrell @leecantrell.bsky.social · 16/02/2026

This is a nice study. I’d anticipate conclusions are a bit biosample specific with respect to reasonable comparable range. (LoD/LoQ dist related). An advantage of protein norm is reduction of a variable in case of other systematic, independent variable. Protein norm has its issues too though.

Lee Cantrell @leecantrell.bsky.social · 31/01/2026

Not to be lost - this is an assessment for discovery experiments to max. biological yield. Targeted will still benefit from added DPPP in many cases

Lee Cantrell @leecantrell.bsky.social · 31/01/2026

I think the reality is that the model in 2020 paper is an ideal scenario for uniform high SNR peptides. In reality, ideal isn't probable for all acquisition. Greater divergence decreases DPPP : accuracy dependency in model. Low SNR is still very much dependent on DPPP to enable LOQ assessment.

Lee Cantrell @leecantrell.bsky.social · 31/01/2026

Yep! Can recreate the plot, but added detector characteristic, SNR, and peak asymmetry. Esp. SNR impacts the quant. If I drop SNR from the model it looks basically identical to the 2020 MCP paper.

Lee Cantrell @leecantrell.bsky.social · 29/01/2026

Check out my US HUPO poster and soon to come pre-print :) Good chromatography and high SNR peaks will be fine with few DPPP. Low abundance features suffer with few DPPP.

230

Lee Cantrell @leecantrell.bsky.social · 08/11/2025

I'm excited to share new results at #HUPO2025! I’ll be presenting our latest work on a next-generation DIA search and FDR pipeline that enables sensitive, accurate and scalable proteomic analysis — in just a fraction of the time required by current algorithms. 📍 Poster PV.01.009 — Monday

0101

Lee Cantrell @leecantrell.bsky.social · 14/09/2025

Clinical protocols are very tricky to standardize across sites.

030

Reposted by Lee Cantrell

PastelBio @pastelbio.bsky.social · 13/09/2025

Pre-analytical drivers of bias in bead-enriched plasma proteomics | EMBO Molecular Medicine www.embopress.org/do... --- #proteomics #prot-paper

2125

Lee Cantrell @leecantrell.bsky.social · 03/08/2025

Yeah just run a sample every 2 minutes for 5 years with no overhead or down time and a 2M instrument costs 1.5 per sample… Pretend 15 min with 75% up time and it’s 15/sample. Adjust cost relative to purchase price.

Lee Cantrell @leecantrell.bsky.social · 08/07/2025

Isomerization? Especially in deamidation prone peptides with NQ. This does split peaks in LC. Eye lens work has seen this for decades. Doesn’t mean deamidation necessarily. Can generate 4plet or more if including modification. Tris in prep?

Or on high mass? I haven’t seen much for >100ng from the 8600.

Not that I’m aware of - though I’d also be nervous about charge capacity on transmission through the tims device. I’d rather see the 7600 or 8600.

It’s unclear that SLIMS has the reproducibility or resolution to achieve this. Marketing figures aside, peptides don’t always make nice Gaussians. Co-resolving charge states likely isn’t an issue. What I want to see is an honest effort on DIA vs PAMAF with a good DIA TOF instrument.

Rationale seems to be more ions = better data. Astral transmits ~1/200th of ions within a target search space at a time (less the overhead and MS1 times). A 400ms SLIMS separation could reasonably replace half the selectivity of the quad and 4x throughput of qTOF.

Lee Cantrell @leecantrell.bsky.social · 15/06/2025

When I check in every month or so, it’s dry. Much drier than bsky