LARIS: ligand-receptor interactions in spatial data

LARIS — Ligand And Receptor Interaction in Spatial transcriptomics — is a separate package with its own repository and documentation. It is described here because it answers the same question as SCALAR under a different constraint, and choosing between them is the first decision, not an implementation detail.

• Repository: https://github.com/genecell/LARIS
• Documentation: https://genecell.github.io/LARIS/
• Install: pip install laris

What changes when cells have coordinates

SCALAR asks whether a ligand is specific to the sender and the receptor specific to the receiver. It cannot ask whether those two cells were ever near each other, because a dissociated dataset does not know. Every ordered pair of cell types is scored as though contact were possible.

In spatial data that assumption is testable, and it is usually false. Two cell types that both express a complementary pair may sit in different layers and never touch. LARIS uses the coordinates: interaction strength is computed per cell against its spatial neighbours, so a pair only scores where the two partners are actually adjacent.

That shifts what the output is. SCALAR returns one score per (LR pair, sender, receiver). LARIS returns both: a score for each individual cell, and the sender-receiver cell-type summary computed from it. So it is not a trade: the cell-type view SCALAR gives you is still there, with the per-cell resolution underneath it, which is what lets the same result be mapped onto the tissue or tested for spatial variability.

What it produces

• Per-cell interaction strength for each ligand-receptor pair.
• Spatially variable LR pairs — pairs whose interaction is patterned across the tissue rather than uniform.
• Sender-receiver scores at cell-type level, the summary that is directly comparable to a SCALAR result.
• Spatial neighbourhood context — which cell-type compositions a given interaction occurs in.

Which one to use

	SCALAR	LARIS
data	dissociated scRNA-seq / snRNA-seq	spatial (MERFISH, Xenium, Visium, …)
unit of the answer	cell-type pair	individual cell and cell-type pair
what constrains a hit	specificity in both partners	specificity and physical proximity
the output can be	ranked, filtered, plotted	all of that, plus mapped onto tissue

They share the interaction database. The CellChatDB tables that piaso.data.load_lr_database fetches are the same input either method takes, so a pair list curated for one transfers to the other.

The natural pairing is both: SCALAR on a dissociated reference to find which interactions exist in the tissue at all, LARIS on the spatial sample to ask where they happen.

The tutorials

Six worked tutorials live with the package, each on a different platform and a different question. They are the fastest way to see which one your data calls for.

	what it covers
Core pipeline	prepareLRInteraction, runLARIS and the plotting, end to end on Slide-tags tonsil. Start here.
Comparing conditions	compareLARIS across samples and subjects, with volcano plots — Visium myocardial infarction.
Comparing at matched cell states	The matched estimator: a joint embedding first, so a difference is not just a difference in composition. MERFISH gut.
Cytome guide	The on-disk path end to end on the tonsil data — building an LR cytome, streaming COSG, and what it guarantees about reproducibility.
Tissue image overlay	H&E at single-cell resolution and the registration it needs — Xenium Prime 5K human skin. The spatial counterpart of the Xenium overlay here.
Two-variable analysis	Crossed cell-type and region labels, when “which cell type” and “where in the tissue” are separate questions.

The list is kept at LARIS/tutorials.