Rationale and Benchmarks for TopN = 1 (PG.Quantity) vs PG.MaxLFQ

[AnirudhKashyap511]

Hi Vadim,

I had a quick question regarding protein quantification in DIA-NN.

From earlier discussions here, I understand that PG.Quantity uses Top N = 1 peptide per protein group for quantification, and that in some cases Top1 can outperform Top3 or other summarization strategies.

Are there any benchmarks comparing Top1 vs Top3 vs MAXLFQ? While I understand that PG.MaxLFQ is generally considered the most robust, we often observe: More protein identifications with PG.Quantity and Strong correlation between PG.Quantity and PG.MaxLFQ.

Because of this, we’re considering using PG.Quantity for certain downstream tasks and just wanted to better understand the theoretical and benchmarking basis behind Top1 being preferable.

Anirudh

[vdemichev]

Hi Anirudh,

From earlier discussions here, I understand that PG.Quantity uses Top N = 1 peptide per protein group for quantification, and that in some cases Top1 can outperform Top3 or other summarization strategies.

Yes.

Are there any benchmarks comparing Top1 vs Top3 vs MAXLFQ?

Top 3 vs MaxLFQ - definitely, not sure about Top 1.

More protein identifications with PG.Quantity and Strong correlation between PG.Quantity and PG.MaxLFQ.

Logical: they estimate the same entity.

Because of this, we’re considering using PG.Quantity for certain downstream tasks and just wanted to better understand the theoretical and benchmarking basis behind Top1 being preferable.

Full guide for recent DIA-NN versions here: https://github.com/vdemichev/DiaNN?tab=readme-ov-file#quantification

Best,
Vadim