William Stafford Noble

I am deleting my Twitter account. Please follow me on BlueSky (@wnoble.bsky.social).

@pwilmarth @neely615 @andy_compbio @Smith_Chem_Wisc Not sure if this is what you are referring to when mentioning PTMs, but we recently also showed that PTMs don’t seem to significantly affect the validity of FDR control in narrow search, as had been previously suggested (doi.org/10.1101/2023.0…).

@pwilmarth @neely615 @andy_compbio @Smith_Chem_Wisc We recently pointed out some of the issues with searching with variable modifications and how to counter that in your FDR analysis (doi.org/10.1101/2023.0…). That paper also suggests how to deal with neighbors (peptides that share parts of their theoretical spectra).

@pwilmarth @neely615 @andy_compbio @Smith_Chem_Wisc The idea behind TDC is that an incorrect match is equally likely to hit the target or the decoy database. This requires special care in constructing decoys. We don't recommend the reversing the proteins; rather, one should reverse or shuffle the peptides.

@pwilmarth @neely615 @andy_compbio @Smith_Chem_Wisc It sounds like you believe target/decoy FDR estimation has a liberal bias because it doesn't take into account some types of errors.  This is not a bound in the statistical sense.  What are the "non-random sources of peptide sequencing errors" and why are they biased?

@pwilmarth @neely615 @andy_compbio @Smith_Chem_Wisc FDR is defined as the expected rate of false discoveries. So it’s a mean rather than an upper or lower bound. If you want an upper bound, please see our recent JPR tutorial on “Bridging the false discovery gap.” pubs.acs.org/doi/10.1021/ac…

Interested in de novo peptide sequencing from #massspec #proteomics data? Check out our latest #preprint about Casanovo! 👉 doi.org/10.1101/2023.0… That's right folks - it's time for another #tweetorial 🧵 (1/9)

Want to learn more about the state of the art in de novo peptide sequencing at #ASMS2023? Come check out our: - Evening workshop #17 for a Casanovo tutorial on Monday - Talk at the AI session on Wednesday #TeamMassSpec

@Smith_Chem_Wisc @neely615 Friends don’t let friends control FDR at the PSM level.

@neely615 Yeah. Every time @thabangh writes a paper I think I need to stop everything and rewrite the code that computes FDR. I thought that was solved. I guess I am not right about much.

Congratulations to Lincoln Harris on being awarded an NRSA fellowship!

Casanova is the new state of the art in de novo peptide sequencing. biorxiv.org/content/10.110…

Check out Polarbear🐻‍❄️, a method to predict the missing scATAC-seq profiles just based on scRNA-seq, and vice versa. 🐻‍❄️ also derives matching in between scRNA-seq and scATAC-seq profiles, enabling integration of different types of single cell measurements. link.springer.com/chapter/10.100…

Our paper about building spectral libraries from DIA data is officially out @JProteomeRes! Thank you to everyone who worked on this and to our reviewers for their feedback, I'm very excited about the final product.

Obligatory post following a bioRxiv submission. In this work we show that 1) PSM-level FDR estimates in practice are invalid and 2) a simple variant of the peptide-level FDR estimate protocol can yield a significantly greater number of peptide detections

Our featured article: Navigating the pitfalls of applying machine learning in genomics go.nature.com/3D55JuZ #Review by @seawhalen, @jmschreiber91, William S. Noble & Katherine S. Pollard @GladstoneInst @Stanford @UW @uwgenome @UCSF @czbiohub

The state of the art in de novo peptide sequencing has just advanced: meet Casanovo, a transformer model that translates from the sequence of peaks in a spectrum to the sequence of amino acids in a peptide. biorxiv.org/content/10.110…

Not going to lie. Part of the reason I created an account was to promote this preprint of mine. We have created a new score, MS1Connect, that measures the similarity between pairs of mass spectrometry runs solely using MS1 scans and without any peptide identifications. 🧵 1/9

@NorSivaeb @nesvilab Thanks, Ron. Not sure I'd call these "edge cases." But sticktoitiveness is what we're aiming for!

FDR estimation in proteomics is a harder problem that it is appreciated. When developing open search strategies we found that modified peptides, not considered in closed search, lead to inaccurate close search FDR. This work explores related issues in the context of subset search

Excited for another 5 years of studying genome 3D architecture! newsroom.uw.edu/news/nih-funde…