Paul Thomas
@pgtimmune.bsky.social
Division of Vaccine and Infectious Disease, Fred Hutchinson Cancer Center
TCRs, influenza virus, anti-tumor immunity, books, dogs, Venice
TCRs, influenza virus, anti-tumor immunity, books, dogs, Venice
Maybe you’ve identified a novel condition-associated population and you want to see where it falls–is it a conventional epitope specific response? If so, we might be able to tell you the pathogen, the epitope, or the HLA-restriction (or all 3)?
July 14, 2025 at 11:52 PM
Maybe you’ve identified a novel condition-associated population and you want to see where it falls–is it a conventional epitope specific response? If so, we might be able to tell you the pathogen, the epitope, or the HLA-restriction (or all 3)?
For each population we define a TCR motif and GEX profile, curating known natural Treg, NKT, ILTCK and similar populations, and several novel populations that we can isolate with similar resolution. The result displays the breadth of the unconventional T cell kingdom.
July 14, 2025 at 11:52 PM
For each population we define a TCR motif and GEX profile, curating known natural Treg, NKT, ILTCK and similar populations, and several novel populations that we can isolate with similar resolution. The result displays the breadth of the unconventional T cell kingdom.
What do these represent? Lots of known unconventional T cell subsets (MAITs, NKTs, various thymic developmental subsets, KIR+ CD8s, and Tregs) and lots of unknown discrete unconventional populations.
July 14, 2025 at 11:52 PM
What do these represent? Lots of known unconventional T cell subsets (MAITs, NKTs, various thymic developmental subsets, KIR+ CD8s, and Tregs) and lots of unknown discrete unconventional populations.
The idea here is that within GEX space, we look for regions where a subset of neighbors have strongly statistically biased usage of specific TCR amino acids (in no particular order). We find a lot of these neighborhoods! (Over 70K for CD8 & 50K for CD4).
July 14, 2025 at 11:52 PM
The idea here is that within GEX space, we look for regions where a subset of neighbors have strongly statistically biased usage of specific TCR amino acids (in no particular order). We find a lot of these neighborhoods! (Over 70K for CD8 & 50K for CD4).
So that accounts for a big chunk of the conventional memory repertoire for both CD4 and CD8 T cells…but what about the rest of the repertoire? The second major analysis we perform is a GEX neighborhood-based amino acid bias assessment of the TCR.
July 14, 2025 at 11:52 PM
So that accounts for a big chunk of the conventional memory repertoire for both CD4 and CD8 T cells…but what about the rest of the repertoire? The second major analysis we perform is a GEX neighborhood-based amino acid bias assessment of the TCR.
As we described in CoNGA 1.0, these TCR clusters converge in GEX space as well, demonstrating the profound effects of shared priming history. Even more dramatically, distinct epitopes targeting the same pathogen also converge in CoNGA GEX space.
July 14, 2025 at 11:52 PM
As we described in CoNGA 1.0, these TCR clusters converge in GEX space as well, demonstrating the profound effects of shared priming history. Even more dramatically, distinct epitopes targeting the same pathogen also converge in CoNGA GEX space.
For some of these motifs, we have “re-discovered” classic immunodominant epitopes from flu (M1 58) or CMV (pp65 NLV). Many others are novel, though we can often assign an HLA restriction and may have a clue about the pathogen they target…
July 14, 2025 at 11:52 PM
For some of these motifs, we have “re-discovered” classic immunodominant epitopes from flu (M1 58) or CMV (pp65 NLV). Many others are novel, though we can often assign an HLA restriction and may have a clue about the pathogen they target…
First, we perform an extensive “TCR convergence” analysis, finding regions of TCR space spanning individuals enriched for classic epitope-specific TCR motifs. We identify over 2000 such groups, representing the breadth of shared immunodominant responses across humans
July 14, 2025 at 11:52 PM
First, we perform an extensive “TCR convergence” analysis, finding regions of TCR space spanning individuals enriched for classic epitope-specific TCR motifs. We identify over 2000 such groups, representing the breadth of shared immunodominant responses across humans
Previously we released the CoNGA algorithm, linking TCR sequence and GEX to identify structure-function relationships in the T cell repertoire. After extensive curation of a wide array of public data (1900 subjects, 6 million cells) we present metaCoNGA.
July 14, 2025 at 11:52 PM
Previously we released the CoNGA algorithm, linking TCR sequence and GEX to identify structure-function relationships in the T cell repertoire. After extensive curation of a wide array of public data (1900 subjects, 6 million cells) we present metaCoNGA.
TIRTL-seq is (almost) our dream TCR repertoire tech: paired TCRs for clones with >=3 cells/sample, and unpaired alpha/beta for all small clones. Millions of T cells for <$200/sample, & works on longitudinal samples. Protocol/primers available now, code & data released soon.(8/11)
September 23, 2024 at 5:28 PM
TIRTL-seq is (almost) our dream TCR repertoire tech: paired TCRs for clones with >=3 cells/sample, and unpaired alpha/beta for all small clones. Millions of T cells for <$200/sample, & works on longitudinal samples. Protocol/primers available now, code & data released soon.(8/11)
TIRTL-seq generates 100s of biological replicates, measures noise & finds growing clones. But matching receptors to VDJdb surprised us: the largest expansions were for EBV, not COVID! COVID clones went down, but EBV remained stable. Is this how big stable clones form? (7/11)
September 23, 2024 at 5:28 PM
TIRTL-seq generates 100s of biological replicates, measures noise & finds growing clones. But matching receptors to VDJdb surprised us: the largest expansions were for EBV, not COVID! COVID clones went down, but EBV remained stable. Is this how big stable clones form? (7/11)
Having abTCR pairs is nice, but we also want to track clonal expansions/contractions in time. We did bulk 5’RACE TCRb seq with UMIs, 10x Genomics scTCR-seq, and TIRTL-seq on a COVID-19 patient sampled pre-infection, & at acute & convalescent timepoints to compare methods. (6/11)
September 23, 2024 at 5:28 PM
Having abTCR pairs is nice, but we also want to track clonal expansions/contractions in time. We did bulk 5’RACE TCRb seq with UMIs, 10x Genomics scTCR-seq, and TIRTL-seq on a COVID-19 patient sampled pre-infection, & at acute & convalescent timepoints to compare methods. (6/11)
We compare abTCR pairs to 10x single cell TCRseq: big clones are correctly paired by T-SHELL, smaller by MADHYPE or both. >90% of TCRbetas overlapping with 10x scTCR-seq have the same TCRalpha pair (~98% correct pairs if we exclude TCRs with 1 cell in 10x). High precision! (5/11)
September 23, 2024 at 5:28 PM
We compare abTCR pairs to 10x single cell TCRseq: big clones are correctly paired by T-SHELL, smaller by MADHYPE or both. >90% of TCRbetas overlapping with 10x scTCR-seq have the same TCRalpha pair (~98% correct pairs if we exclude TCRs with 1 cell in 10x). High precision! (5/11)
Combinatorial TCR pairing matches TCRa & TCRb by well occurrence patterns. Problem: expanded clones are found in all wells and cannot be paired. We invented a new algorithm (T-SHELL) using a/b read count correlation across wells instead of presence/absence patterns (4/11)
September 23, 2024 at 5:28 PM
Combinatorial TCR pairing matches TCRa & TCRb by well occurrence patterns. Problem: expanded clones are found in all wells and cannot be paired. We invented a new algorithm (T-SHELL) using a/b read count correlation across wells instead of presence/absence patterns (4/11)
Inspired by tiny volumes of SS3xpress @sandberglab, combinatorial TCR pairing suggested by @bryan_howie & @MEBirnbaum Bayesian inference method MADHYPE reimplemented for GPU. We process 10 mln cells in 384 wells for <$200 & scale to 30 mln cells & 1 mln TCRabs in 2300 wells (3/11)
September 23, 2024 at 5:28 PM
Inspired by tiny volumes of SS3xpress @sandberglab, combinatorial TCR pairing suggested by @bryan_howie & @MEBirnbaum Bayesian inference method MADHYPE reimplemented for GPU. We process 10 mln cells in 384 wells for <$200 & scale to 30 mln cells & 1 mln TCRabs in 2300 wells (3/11)
TIRTL-seq is (almost) our dream TCR repertoire tech: paired TCRs for clones with >=3 cells/sample, and unpaired alpha/beta for all small clones. Millions of T cells for <$200/sample, & works on longitudinal samples. Protocol/primers available now, code & data released soon.(8/11)
September 23, 2024 at 5:25 PM
TIRTL-seq is (almost) our dream TCR repertoire tech: paired TCRs for clones with >=3 cells/sample, and unpaired alpha/beta for all small clones. Millions of T cells for <$200/sample, & works on longitudinal samples. Protocol/primers available now, code & data released soon.(8/11)
TIRTL-seq generates 100s of biological replicates, measures noise & finds growing clones. But matching receptors to VDJdb surprised us: the largest expansions were for EBV, not COVID! COVID clones went down, but EBV remained stable. Is this how big stable clones form? (7/11)
September 23, 2024 at 5:25 PM
TIRTL-seq generates 100s of biological replicates, measures noise & finds growing clones. But matching receptors to VDJdb surprised us: the largest expansions were for EBV, not COVID! COVID clones went down, but EBV remained stable. Is this how big stable clones form? (7/11)
Having abTCR pairs is nice, but we also want to track clonal expansions/contractions in time. We did bulk 5’RACE TCRb seq with UMIs, 10x Genomics scTCR-seq, and TIRTL-seq on a COVID-19 patient sampled pre-infection, & at acute & convalescent timepoints to compare methods. (6/11)
September 23, 2024 at 5:25 PM
Having abTCR pairs is nice, but we also want to track clonal expansions/contractions in time. We did bulk 5’RACE TCRb seq with UMIs, 10x Genomics scTCR-seq, and TIRTL-seq on a COVID-19 patient sampled pre-infection, & at acute & convalescent timepoints to compare methods. (6/11)
We compare abTCR pairs to 10x single cell TCRseq: big clones are correctly paired by T-SHELL, smaller by MADHYPE or both. >90% of TCRbetas overlapping with 10x scTCR-seq have the same TCRalpha pair (~98% correct pairs if we exclude TCRs with 1 cell in 10x). High precision! (5/11)
September 23, 2024 at 5:25 PM
We compare abTCR pairs to 10x single cell TCRseq: big clones are correctly paired by T-SHELL, smaller by MADHYPE or both. >90% of TCRbetas overlapping with 10x scTCR-seq have the same TCRalpha pair (~98% correct pairs if we exclude TCRs with 1 cell in 10x). High precision! (5/11)
Combinatorial TCR pairing matches TCRa & TCRb by well occurrence patterns. Problem: expanded clones are found in all wells and cannot be paired. We invented a new algorithm (T-SHELL) using a/b read count correlation across wells instead of presence/absence patterns (4/11)
September 23, 2024 at 5:25 PM
Combinatorial TCR pairing matches TCRa & TCRb by well occurrence patterns. Problem: expanded clones are found in all wells and cannot be paired. We invented a new algorithm (T-SHELL) using a/b read count correlation across wells instead of presence/absence patterns (4/11)
Inspired by tiny volumes of SS3xpress @sandberglab, combinatorial TCR pairing suggested by @bryan_howie & @MEBirnbaum Bayesian inference method MADHYPE reimplemented for GPU. We process 10 mln cells in 384 wells for <$200 & scale to 30 mln cells & 1 mln TCRabs in 2300 wells (3/11)
September 23, 2024 at 5:25 PM
Inspired by tiny volumes of SS3xpress @sandberglab, combinatorial TCR pairing suggested by @bryan_howie & @MEBirnbaum Bayesian inference method MADHYPE reimplemented for GPU. We process 10 mln cells in 384 wells for <$200 & scale to 30 mln cells & 1 mln TCRabs in 2300 wells (3/11)