🔥Telegram İfşa

In a new work led by Rebekah Loving, joint with Ali Mortazavi and his lab, we introduce a method based on kallisto for accurate quantification of Oxford Nanopore or PacBio long-read RNA-seq, and demonstrate its accuracy with numerous comparisons and benchmarks 1/🧵 biorxiv.org/content/10.110…

Long-read sequencing has been transformative for genome sequencing and transcript (isoform) discovery, but read depth, constrained by cost, has limited its use for quantification. That has changed in the past few years. And error rates have dropped. 2/ x.com/geertvangeest/…

We decided to adapt kallisto, which Delaney Sullivan has recently updated and improved in many ways, for long-read quantification. Rebekah Loving figured out that several modifications and improvement, especially w.r.t fragment lengths were needed. x.com/lpachter/statu… 3/

We needed a controlled dataset to benchmark on, so the Ali Mortazavi lab generated the ideal testbed: Illumina and Oxford Nanopore sequenced libraries with and without exome capture (to yield more reads from exons). And of course biological replicates for each. 4/

The main result (Fig. 1) is that long-read quantification with lr-kallisto is consistent with kallisto short-read quantification. The CCC (concordance correlation coefficients) are extremely high, demonstrating both high quality data and high accuracy quantification 5/

We benchmarked some other tools that have recently been developed for long-read quantification, including Bambu, Oarfish, and IsoQuant. lr-kallisto is faster and more accurate. 6/

We also performed extensive benchmarking on the amazing #LRGASP resource that was recently published... x.com/anaconesa/stat… 7/

... and we performed numerous benchmarks on simulated data. These benchmarks corroborated our results on experimental data, and provided information on performance limits. In particular, lr-kallisto performance starts to degrade at very high error rates. 8/

We also tested lr-kallisto on the main dataset used in the recent Oarfish preprint (the HCT116 cell line w/ Illumina and ONT), and succeeded in replicating the Oarfish results while showing that lr-kallisto outperforms Oarfish. 9/

lr-kallisto also performed well on a PacBio dataset which was Illumina sequenced as well, albeit with less performance gain because the long-read data has a very high error rate (12.4%). 10/

The preprint has detailed analyses not just demonstrating performance but also examining why lr-kallisto performs well and how it can be optimized. Tl;dr, longer k-mers in the transcriptome de Bruijn graph can be helpful. 11/

Many thanks to Barbara Wold, who also supervised the work, and the team from Ali Mortazavi lab team from UC Irvine who did exceptional work generating the matched ONT and Illumina data. An incredible accomplishment by Rebekah Loving! x.com/MarissaKawehi/… 12/

Code to reproduce results is available at github.com/pachterlab/LSR… 13/13

The mantra that spatial transcriptomics is about location, location, location is catchy, but what does it really mean? We have just posted biorxiv.org/content/10.110…, work of Kayla Jackson et al., that describes the concordex method for identifying spatial homogeneous regions. 1/🧵

biVI models the biophysical processes generating nascent and mature single-cell transcriptomes using variational autoencoders. Maria Carilli Gennady Gorin Yongin Choi Lior Pachter nature.com/articles/s4159…

We are excited to welcome Dr. Lambda Moses, our new Postdoctoral Research Scientist! She specializes in spatial-omics data analysis and software development. In the bianca dumitrascu group, Lambda focuses on computational biology, single-cell genomics, and statistical bioinformatics.

I'm giving a talk this Thursday at 5p at Harvard Medical School in the Dept. of Biomedical Informatics. If you are in the area and would like to attend in person- please DM me.

☛ scRNA-seq is solved, we're moving on to multiomics and ML ☛ forgot to mention, DE for sc is just DE for bulk, we don't have a way to use cell-level info ☛ also DE for bulk requires normalization and every method gives different results; try a bunch and see if they agree

This is legit.

Watch the talk from Joseph Rich, grad student in Lior Pachter's lab, from BioC2024 timestamp ~12:30 youtu.be/T6cOyKHlxpE?si…