Science1 distinct publisher3 min readPublished
The method swaps curve-fitting for a Lomb-Scargle periodogram, so irregular pseudotime and branching trajectories need no branch assignments, at the cost of a ranked gene list that cannot say which lineage diverged.
The Scientist · Science desk
Compiled by The ScientistSomething wrong?How this is made
The curve-fitting step is where the trouble starts. Conventional trajectory analysis writes gene expression as a smooth function of pseudotime, which presumes the shape you are hunting for, and it needs cells assigned to branches before it can fit anything at all [5]. scLS, reported in Nucleic Acids Research, moves the description into the frequency domain, where the spacing of samples along pseudotime stops being an obstacle [7][8].
The dynamic test scans a predefined grid of frequencies, evaluates a false-alarm probability at each, and takes the smallest as the gene's P-value [9]. Minimum-of-many statistics deserve a second look, because the more frequencies you scan, the smaller that minimum gets by chance. A false-alarm probability is built to carry exactly that correction; the interesting question is how well it carries it when cells pile up unevenly along an ordering that was itself estimated. The account available quotes no calibration figures either way [15].
The shifted test is the more novel half. Each condition gets its own periodogram, which is two transforms per gene [16], and the statistic is the distance between the two power spectra, converted to a right-tailed P-value under a normal approximation [10]. That approximation is where I would look first. Screening a transcriptome puts every decision deep in the tail of the null distribution, and a normal approximation that behaves beautifully near the middle can be off by a lot out there.
Competitive detection at lower computational cost, on simulations and on real data [11], is the right shape of evidence for a first-pass screen, though it says nothing about what happens on your own cluster. The measurement that governs a million-cell atlas is peak memory and wall clock at a million cells, and that is a different quantity from detection power in a simulation with a known ground truth.
What this leaves open is whether the ordering itself deserved the trust. Single-cell RNA sequencing does not follow one cell through time; pseudotime is an arrangement of snapshots computed after the fact [1][2], and a downstream test inherits whatever that arrangement got wrong. A gene flagged as dynamic over pseudotime is judged by its position in that arrangement, not by any individual cell observed changing over time. Hamada names disease progression and drug response among the intended applications [14], which are settings where a shift between conditions is often the effect of interest and a smooth trend is not. So, a view with its condition attached: used the way the authors position it, as an efficient screen that prioritises genes ahead of lineage-aware analysis [12], scLS earns a slot in the pipeline, provided the ranked list travels with the fact that its ranking is conditional on a trajectory inferred upstream.
Ranked by verification strength, evidence, and original report placement.
Single-cell RNA sequencing measures gene expression in individual cells rather than averaging across millions of cells, but it does not track the same cell continuously over time.
Trajectory inference approaches computationally arrange cellular snapshots along an inferred developmental trajectory known as pseudotime.
Recent advances in trajectory inference have enabled data sets containing hundreds of thousands or even millions of cells, but downstream analyses of these trajectories, particularly identification of differentially expressed genes, remain challenging.
Differentially expressed genes can show two patterns: expression that changes dynamically over pseudotime, and patterns shifted between two conditions. Dynamic expression analysis is more widely used, though identifying shifted patterns is also important.
Cells can be distributed irregularly along pseudotime and inferred trajectories may contain multiple branches, for which conventional models require explicit regression models or branch assignments; conventional methods typically fit gene expression as a smooth function of pseudotime.
A team led by Assistant Professor Hitoshi Iuchi and Professor Michiaki Hamada of the Faculty of Science and Engineering at Waseda University in Japan developed a downstream analysis algorithm called scLS, a computational method for identifying pseudotime-associated genes from single-cell RNA sequencing data.
Distinct publishers with included, body-backed reporting in this cluster.
phys.org
1 article · September 1, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
science
Basel's tree-shaped embedding argues your flat single-cell maps have been showing you artifacts1 distinct publisher
science
Malva indexes public single-cell archives by nucleotide sequence instead of by gene1 distinct publisher
science
Columbia says depression stalls new neurons. The subtype it implies has no test yet2 distinct publishers
science
A rare neuron-like cell may spare dogs a jaw resection1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Peer-reviewed paper, single relayed account
The mechanism is described precisely enough to be criticised — frequency grid, minimum false-alarm probability, spectral distance, normal approximation — and it sits behind a Nucleic Acids Research citation with a DOI. But everything a reader can see comes through one Phys.org retelling of the lab's own announcement, and the comparative result that would justify adopting the method has no baseline, data set or number attached.
A paper, and nothing after it
A journal publication is an announcement, not uptake. Nobody in this reporting has installed scLS, cited it, or run it on their own data; there is no repository, download count, or named group using it. We will not turn a DOI into an adoption number.
Restrained claims, still unmeasured
This is unusually disciplined for a university announcement: it says outright that scLS cannot localise dynamics to a lineage and casts the tool as a first pass, not a replacement. The small overhang comes from two habits — 'competitive performance' and 'more computationally efficient' with nothing quantified, and Hamada's sweep from differentiation to drug response without one worked example behind any item on the list.
The lab describing its own tool
Every evaluative sentence here originates with the people who wrote the method. Both quoted voices are the two lead authors, the framing of what was hard before and what is easier now is theirs, and the citation block reads like the appendix of an institutional release. That is not a reason to disbelieve the mechanism, which is specific and checkable; it is a reason to treat the verdict on speed and accuracy as advocacy until an outside group reproduces it.
Solid on what it is, thin on how well it works
We would defend the description of scLS — the statistics are laid out in enough detail that a specialist could argue with them, and the stated inability to name a branch is a real constraint on how it can be used. We would not yet defend any statement about how it performs, and the total absence of adoption or software signals leaves half the picture blank.