Kevin Coombes

Kevin Coombes

@krc77.bsky.social

Husband, father, grandfather, cyclist, tifoso, mathematician, biostatistician, bioinformatician. Reader of books, watcher of movies, listener to music, wanderer through museums. Appreciator of foreign travel.

77 followers162 following214 posts

Kevin Coombes
@krc77.bsky.social

Yesterday, I turned 71. Today, my wife and I climbed (on foot) along the porticos from the Arco de Meloncello in Bologna up to the Monisterio di San Luca. Picked out a place to stand during Saturday's Giro dell'Emiglia.

Kevin Coombes
@krc77.bsky.social

A former postdoc asked me about "AI scientists", saying they use demographic, clinical, laboratory, and socioeconomic variables to split the same sample set multiple ways. Then perform t-tests on the same gene expression data for each split. How to adjust for multiple testing? #stats #AI 1/n

View full thread
Kevin Coombes
@krc77.bsky.social

Another approach is to first fit a comprehensive linear model using all splitting variables as predictors. This idea suggests that you might need to consider all n-choose-2 pairs of interactions between predictors. Might also look at what adding one predictor does on top of everything else. 4/n

Kevin Coombes
@krc77.bsky.social

Alternatively, perhaps one should start with a latent class analysis of the potential splitting variables on the samples. And then fit linear models with the classes as predictors of gene expression. 5/n

Kevin Coombes
@krc77.bsky.social

A former postdoc asked me about "AI scientists", saying they use demographic, clinical, laboratory, and socioeconomic variables to split the same sample set multiple ways. Then perform t-tests on the same gene expression data for each split. How to adjust for multiple testing? #stats #AI 1/n

Kevin Coombes
@krc77.bsky.social

He also says that people tell him it's enough to report the results of one such t-test analysis, with FDR adjustments across genes, while ignoring the other analyses. That's clearly wrong, even if you are only looking at the expression of just one gene. 2/n

Kevin Coombes
@krc77.bsky.social

One approach would be to perform a Bonferroni adjustment across splits. That doesn't account for the lack of independence between splits. For example, consider gene expression in lung cancer. Split on smoking status, sex, race, urban or rural, etc. Many of those are related. 3/n

Kevin Coombes
@krc77.bsky.social

Title: PLASMA: Partial LeAst Squares for Multiomics Analysis PubMed: pubmed.ncbi.nlm.nih.gov/39858069/ Journal: www.mdpi.com/2072-6694/17... Git: gitlab.com/krcoombes/pl... CRAN: cran.r-project.org/package=plasma

pubmed.ncbi.nlm.nih.govPLASMA: Partial LeAst Squares for Multiomics Analysis - PubMed<span><b>Background/Objectives</b>: Recent growth in the number and applications of high-throughput "omics" technologies has created a need for better methods to integrate multiomics data. Much progress has been made in developing unsupervised methods, but supervised methods have lagged behind. <b>Methods</b></span> …
Kevin Coombes
@krc77.bsky.social

Title: SillyPutty: Improved clustering by optimizing the silhouette width PubMed: pubmed.ncbi.nlm.nih.gov/38848330/ Journal: journals.plos.org/plosone/arti... Git: gitlab.com/krcoombes/si... CRAN: cran.r-project.org/package=Sill...

pubmed.ncbi.nlm.nih.govSillyPutty: Improved clustering by optimizing the silhouette width - PubMedClustering is an important task in biomedical science, and it is widely believed that different data sets are best clustered using different algorithms. When choosing between clustering algorithms on the same data set, reseachers typically rely on global measures of quality, such as the mean silhoue …
Kevin Coombes
@krc77.bsky.social

Topological Structures in the Space of Treatment-Naïve Patients with Chronic Lymphocytic Leukemia Journal: www.mdpi.com/2072-6694/16... PubMed: pubmed.ncbi.nlm.nih.gov/39123390/ Git: gitlab.com/krcoombes/cl... CRAN: cran.r-project.org/package=RPoi...

pubmed.ncbi.nlm.nih.govTopological Structures in the Space of Treatment-Naïve Patients with Chronic Lymphocytic Leukemia - PubMedPatients are complex and heterogeneous; clinical data sets are complicated by noise, missing data, and the presence of mixed-type data. Using such data sets requires understanding the high-dimensional "space of patients", composed of all measurements that define all relevant phenotypes. The current …
Kevin Coombes
@krc77.bsky.social

Cycling GOATs. Score for measuring cycling greatness: Nine events (3 GT, 5 monuments, road world champ). One point for each win; one point for each kind of event ever won. Top 10: Merckx, 42 Pogacar, 22 Hinault, 21 Coppi, 21 Binda, 18 Gimondi, 17 Van Looy, 16 De Vlaemink, 16 Bartali, 16 Kelly, 15

Kevin Coombes
@krc77.bsky.social

Is there a setting to get Bluesky feeds to move forward in time? Default is to start with newest posts and move backwards. So, you see responses before original posts. I'd like to pick up the steam where I last left it, and move forward.

Kevin Coombes
@krc77.bsky.social

I have a theory that cycling commentators don't actually watch the race. Or maybe it's only Bob Roll and Christian Vande Velde. Stage 3 of the TdF, they didn't mention David Gaudu until he crossed the line. Stage 4, the same with Ben Turner. They only look for things that confirm their expectations.

Kevin Coombes
@krc77.bsky.social

#stats question: I have data w/3 categorical variables: 3-level copy number (CN) of a gene, 4-level tumor stage, 7-level histology. Both stage and histology are related to copy number, and to each other. Best model/test to see if stage or histo is a more direct driver of CN?

Kevin Coombes
@krc77.bsky.social

@jezcox.bsky.social, I am just watching the replay of Tro Bro Leon. Heard you talking about Breton teams in green. Europcar was one of those in the early 2010's. Started as Bonjour; now TotalEnergies. A couple of years ago, B&B Hotels-KTM was in green and the "team de glaz".

Kevin Coombes
@krc77.bsky.social

@jezcox.bsky.social , while I enjoy your commentary, the mispronunciation of Italian names is grating. "C" before E or I sounds like English ch. That's true of cello and Marcellusi. If you put H between the C and the vowel, it is pronounced hard, like a k. This is true of Chieti and Vecchio.

You are up to date.