Understanding COMMOT
Recommended Reading: 【Bioinformatics】 Bioinformatics Analysis Index
1. Overview
2. Theory
3. Results
4. Discussion
Cang, Z. et al. (2023). Nature methods, 20(2), 218-228. (ref)
1. Overview
⑴ CCC (cell-cell communication)
① Important for understanding biological heterogeneity: Cell fate determination, cell differentiation patterns, pathological mechanisms
② Various forms: Ligand-receptor interactions, cell-cell interactions, organ-level interactions (e.g. insulin), organism-level interactions (e.g. pheromones)
③ Clinical biomarkers: ICI (immune-checkpoint inhibitor), tissue regeneration research
○ Example: Significant increase in necroptosis-related hepatocyte and macrophage interactions in APAP-induced AKI (acute liver injury) tissue
Figure 1. Example of tissue regeneration research
⑵ Existing CCC Analysis Algorithms
① Bulk RNA-seq based: BulkSignalR, squidpy, IPA, Omnipath
② scRNA-seq based: CellTalkDB, CellPhoneDB, CellChat, ICELLNET, NicheNet, SoptSC, CytoTalk, scTensor, CCCExplorer, Connectome, Ramilowski, FlowSig, scSeqComm
③ Spatial Transcriptomics (ST) based: Giotto, spata2, CellPhoneDB v3, stLearn, SVCA, MISTy, NCEM, COMMOT, SCOTIA, STopover, cytosignal, SpatialDM, SpaTalk, stMLnet, HoloNet, DeepLinc
Figure 2. Examples of CCC analysis algorithms using ST
○ Giotto: Constructs spatial proximity graph to identify interactions via membrane-bound ligand-receptor pairs
○ CellPhoneDB v3: Restricts interactions to cell clusters within the same spatially-defined microenvironment
○ stLearn: Associates co-expression of ligand and receptor genes with spatial diversity of cell types
⑶ COMMOT
① Necessity: Existing methods cannot handle multiple-species interaction
○ Multiple ligand types can bind multiple receptor types, leading to competition
○ 72% of ligands and 60% of receptors bind to multiple types
Figure 3. Competition simulation due to multi-species interaction
② Need for OT (optimal transport)
○ OT is defined as the method of optimally allocating resources from origin (supply) to destination (demand)
○ This concept has been studied in economics, physics, astronomy, computer science, etc.
○ Multi-species interaction requires numerical analysis, and the process of distributing gene expression from ligands to receptors is similar, hence OT is needed
Figure 4. OT analogy with moving sandcastles
2. Theory
⑴ Problem Definition of OT
Figure 5. Problem Definition of OT
① Can be used for interaction between one ligand and one receptor
② To apply to multi-species interaction, the problem needs to be reformulated as COT
⑵ Problem Definition of COT (collective OT)
Figure 6. Problem Definition of COT
① Transport plan P ∈ ℝ+nl×nr×ns×ns
○ nl: Number of ligands
○ nr: Number of receptors
○ ns: Number of spots or cells (in case of single cell ST)
② Ideally, ∑j ∑l P(i,j)(k,l) = αi(k), ∑i ∑k P(i,j)(k,l) = βj(l)
○ However, in noisy data this assumption may not hold
○ Instead, define unexplained ligand and receptor quantities μ, ν and solve optimization by panelizing them
⑶ Summary of COT Problem Solving
⑷ Step 1. Apply spatial cutoff to cost function C to consider only spatially proximal interactions as valid
Figure 7. Cost function using spatial information
⑸ Step 2. Express F(·) term as Shannon entropy term and L2 regularizer term
⑹ Step 3. Apply Lagrange multiplier method
① Principle of Lagrange multipliers
Figure 8. Principle of Lagrange multipliers
② Example of Lagrange multipliers
Figure 9. Example of Lagrange multipliers
③ Step 3-1. Introduce additional variables for linearity in Lagrange formulation
④ Step 3-2. First application of Lagrange method: Apply to 5 out of 9 variables first as it’s hard to apply all at once
⑤ Step 3-3. Second application of Lagrange method: Apply to remaining 4 variables
⑥ Step 3-4. Express complex equation as update formula
⑦ Step 3-5. Final conclusion
⑧ Numerical process to solve complex equation: Uniqueness or convergence of the solution is not guaranteed
Figure 10. Numerical solving of complex equation
3. Results
⑴ Visium data shows typical skin structure
① Mainly shows the lower epidermis and dermis part of the skin, which is composed of epidermis, dermis, and smooth muscle
② The inner dermal part being basal cell aligns with expectations
③ As seen from pseudotime results, basal cells divide and produce other cells
Figure 11. Visium skin data
⑵ Noteworthy Interactions
① GAS6-TYRO3 interaction inferred by COMMOT
② Interaction strength for ligand i, receptor j, and spot l is calculated as follows: Result
**Figure 12. **Visualization of GAS6-TYRO3 interaction strength
③ Interaction strength between clusters is calculated as follows
Figure 13. Cluster-level GAS6-TYRO3 interaction strength
④ GAS6-TYRO3 interaction experimentally validated
Figure 14. Experimental validation of GAS6-TYRO3
⑤ GAS-TYRO3 interaction is known as an innate immune suppressor and may relate to the role of unconventional T cells (e.g. γδ T cells, MAIT cells) in skin tissue homeostasis and local immune response
⑶ Noteworthy Pathways
① Interaction direction: Defined as vector sum of relevant transport plan entries limited to ligand-receptor pairs of major pathways. WNT and TGF-β have similar directions due to antagonistic roles in basal cell proliferation
Figure 15. Interaction directions by major pathways
② Basal cell markers (KRT15, KRT5) and granular cell markers (LOR, FLG) related to basal cell proliferation are prominently observed in signaling pathways like WNT and TGF-β
**Figure 16. **Interaction strength of major pathways and gene association analysis
⑷ Noteworthy Visualizations
**Figure 17. **Other noteworthy visualizations
Figure 18. Other noteworthy visualizations
(Spot clustering via spot embedding)
4. Discussion
⑴ Strengths
① Enables analysis of multi-species interaction
② Provides various downstream analyses
③ Validated across five ST platforms and various tissues
④ Runs on Python and is relatively easy to use
⑵ Limitations
① The 1000 μm used as spatial cutoff in cost function is biologically questionable
② Convergence and uniqueness of the COT solution are not sufficiently validated
③ Protein abundance and modifications are not included in ligand-receptor interactions
④ Cannot analyze triple interactions
Figure 19. Triple interaction of FCER2A, ITGAX, ITGB2 using AlphaFold3
Input: 2025.04.05 08:33