An important example involves 10x Genomics Chromium protocols. Some protocols capture RNA transcripts primarily from the 3′ end, while newer approaches can capture transcripts from the 5′ end.

That difference might appear relatively minor, but it changes how sequencing reads are oriented relative to the original RNA transcript.

The researchers found that velocyto systematically assumes the opposite transcript direction compared with 10x Genomics Cell Ranger when processing 5′ sequencing data.

Incorrect counts can change biological conclusions

RNA velocity depends heavily on correctly determining whether individual RNA molecules represent spliced or unspliced transcripts.

When the researchers compared velocyto with other approaches, the orientation problem in 5′ data resulted in different transcript assignments. Those differences then affected the RNA velocities calculated from the data.

Importantly, the problem did not stop at a technical disagreement between software programs. Changes in the calculated RNA velocities could lead researchers to different interpretations of how cells were changing over time.

This demonstrates why the initial quantification step is critical. If spliced and unspliced molecules are incorrectly classified, the resulting model of cellular trajectories may also be inaccurate.

Introducing tidesurf

To address these limitations, the researchers developed tidesurf, a command-line tool for quantifying spliced and unspliced RNA molecules.

Overview of the tidesurf algorithm

(A) Tidesurf takes a GTF file with transcript annotations and the Cell Ranger output, containing the alignments in BAM format, as input and produces count matrices for spliced, unspliced, and ambiguous mRNA molecules. (B) The annotated genome is used to construct an index. To this end, a list of intervals represented by their genomic starting position is constructed, where for each interval, the set of overlapping transcripts is stored. (C-F) Examples of reads for the different read types used in tidesurf. Two example transcripts with different final exon compositions are shown. Exon reads overlap with exons for all transcripts for a gene (C), while exon-exon reads span junctions between exons (D). Intron reads are those aligned to introns for all transcripts (E), while ambiguous reads map to exons for some of the transcripts, and to introns for others (F).

Unlike velocyto, tidesurf was designed to account for both 3′ and 5′ single-cell RNA sequencing protocols. This allows researchers to use a consistent quantification approach across different experimental designs.

The team evaluated tidesurf using four publicly available 10x Genomics Chromium datasets and compared its results with other current quantification approaches.

For conventional 3′ sequencing data, where velocyto generally performs as expected, tidesurf reliably recovered the appropriate transcript counts.

More importantly, tidesurf also produced reliable counts for 5′ sequencing data, where the researchers identified unexpected results from velocyto.

Improving confidence in RNA velocity

RNA velocity can provide insights that are difficult to obtain from a static gene expression profile. Instead of simply identifying what type of cell is present, researchers can investigate whether cells appear to be moving toward particular developmental or functional states.

That makes accurate splicing-state quantification especially important. Errors introduced early in the analysis can propagate through the RNA velocity workflow and ultimately affect biological conclusions.

The findings indicate that researchers working with 5′ single-cell RNA sequencing data should carefully consider which quantification software they use. The authors specifically advise against continued use of velocyto for 5′ sequencing data because of the transcript orientation issue they identified.

Tidesurf provides an updated alternative designed around both major 10x Genomics sequencing orientations. By improving the reliability of spliced and unspliced transcript measurements, it could help researchers generate more dependable RNA velocity estimates across a wider range of single-cell RNA sequencing experiments.

Availability – A Python implementation of tidesurf is available from PyPI and at https://github.com/janschleicher/tidesurf