Check each claim against its experiments, reported results, scope, and stated limitations.Know what the science actually supports before you trust the answer.
Press Enter β΅ to start review
Explore by Goal
"The most beautiful experience we can have is the mysterious. It is the fundamental emotion that stands at the cradle of true art and true science."
- Albert Einstein
Quick Explanation
Copied
Totem (Smolander et al., 2023) in one line
Totem tackles trajectory-inference parameter sensitivity by generating many k-medoids clusterings, building an MST for each, and using a cell-connectivity criterion (MST edge-degree structure) to guide which tree-shaped trajectories should be smoothed and exported as pseudotime lineages.
Long Explanation
Paper Review (critical + skeptical): Cell-connectivity-guided trajectory inference from single-cell data
Target paper:
1) Visual: How Totemβs βconnectivity-guidedβ selection works
Core mathematical object: Totem computes connectivity from each clusteringβs MST topology, where cluster-vertex degree is normalized by the number of vertices, and then builds a cell-level connectivity profile by averaging these connectivity vectors across the ensemble of MSTs.
The paper reports that Totem is statistically better on nonlinear tree trajectories, while linear trajectories are strongest for Slingshot (with Totem comparable to TinGa).
Critical note: This figure intentionally encodes only what the paper explicitly states (directional superiority/comparability), not magnitude across all datasets.
3) Methodological critique: where Totem is strong
Reduces user tuning burden: Totem replaces manual parameter hunting with an ensemble-and-selection workflow (generate many clusterings; rank candidates using connectivity structure; then smooth).
Topology-aware guidance: Instead of using only cluster compactness/separation on reduced data (e.g., ASW/VRC directly on embeddings), Totem constructs a topology-level signal (MST edge-degree) and converts it to a cell-level connectivity gradient.
Benchmarking is aligned to the problem: The dynverse benchmarking framework used in the paper evaluates both topology (HIM) and cell assignment on branches (F1 branches), which matches Totemβs stated bottleneck: MST topology is sensitive and often wrong.
Tree-shaped restriction: Totem explicitly cannot infer trajectories with cycles or simultaneous divergence+convergence; it also does not handle disconnected trajectories in the current implementation.
Dependence on upstream preprocessing: Totem assumes input normalization/QC and depends on dimensionality reduction/embedding and feature extraction choices. The paper emphasizes that upstream choices affect identifiable cell types and thus trajectories.
Ensemble size is a tuning knob in disguise: Robustness is studied for number of clustering results and k-range. Even if robustness improves with more clusterings, computational cost and residual sensitivity remain practical concerns.
Potential overfitting to the evaluation metric: Totem selects trajectories to be better on dynverse benchmarks; if the connectivity criterion correlates with benchmark topology/fate definitions, performance may not translate to real biological processes with different ground-truth notions of milestones. The paper does recommend marker-based biological validation, but the mechanism of βwhat the connectivity captures biologicallyβ remains indirect.
The paper compares four clustering ranking strategies when building MSTs + trajectories: ASW, VRC, Random, and Totem (cell-connectivity + VRC). It distinguishes two regimes: (i) selecting the single best trajectory among top-ranked clusterings, and (ii) averaging performance across multiple top-ranked trajectories.
Skeptical interpretation: Random doing well only when you allow yourself to pick the single best candidate is not βproofβ that connectivity is wrong; itβs consistent with ensemble diversity. The value of Totem is therefore strongest when users want an automatic selection that preserves good average performance without relying on manual marker-heavy iteration.
6) Reproducibility & implementability (from the paper text provided)
Open-source availability: Totem is available as an R package on GitHub.
Benchmark artifacts: The benchmark data are stated to be available on Zenodo; benchmarking code is stated to be on a GitHub repository.
Determinism caveat: The ensemble clustering uses CLARA with RNG enabled; therefore results can vary with random seeds unless the user controls RNG seeds.
7) What would disprove Totemβs main claim?
If connectivity-guided selection repeatedly fails to recover correct milestone topology in real datasets where ground-truth exists at the cluster/milestone level (e.g., externally validated differentiation order), the core connectivity rationale would be weakened. (The paper itself only benchmarks in the dynverse framework and recommends biological marker validation.)
If, after controlling upstream embedding/preprocessing, Totemβs connectivity criterion no longer improves over VRC/ASW baselines for nonlinear trajectories, then the connectivity signal is not genuinely topology-informative (it would be redundant).
Author-review deep links
Feedback:
Updated: April 30, 2026
BGPT Paper Review
Study Novelty
90%
Novelty is high because Totem adds a new ensemble-based selection criterion (cell-connectivity derived from MST topology) specifically designed to guide which tree-shaped trajectories to smooth, rather than only tuning clustering hyperparameters or relying on embedding-based clustering scores.
Scientific Quality
80%
Quality is solid: clear algorithmic workflow, explicit connectivity definition, and benchmarking comparisons against Slingshot/TinGa and clustering selection baselines with multiple evaluation metrics are described. Skeptical red flags are mainly about external validity beyond dynverse and the indirect biological nature of connectivity; the tree-shaped limitation and upstream preprocessing dependence remain central constraints. Evidence strength is moderate-to-strong because the provided paper text includes method and result descriptions, but not all underlying benchmark numeric distributions.
Study Generality
60%
Generality is moderate: the method targets tree-shaped trajectory topologies and inherits key assumptions from Slingshot (embedding distances/topology from MST). It may be less general for cyclic or disconnected processes, and performance likely depends on embedding/preprocessing choices emphasized by the authors.
Study Usefulness
80%
Practically useful for users who want a more automatic, topology-aware clustering selection workflow for nonlinear tree-shaped differentiation trajectories, with built-in visualization of connectivity gradients/milestones. Its impact is constrained by its topology limitations and required upstream preprocessing decisions.
Study Reproducibility
70%
Reproducibility is fairly strong: the paper states open-source availability for Totem, and indicates benchmark data/code availability. However, the ensemble clustering uses RNG seeds (not fully specified here), and robust replication requires careful control of upstream preprocessing and random seeds.
Explanatory Depth
70%
Explanatory depth is moderate: connectivity provides a plausible topology-level signal for milestone placement, but the paper does not fully mechanistically prove that MST-degree-derived connectivity corresponds to specific biological transition rates; rather, it is validated computationally and via recommended marker checks.
Not applicable: this review request does not provide raw per-cell inputs, connectivity vectors, or clustering/MST outputs needed to run a faithful computation pipeline from the paper.
Get emailed when your analysis is done!
We'll email you the results when your analysis is finished.
Hypothesis Graveyard
Totemβs connectivity advantage is purely an artifact of selecting among many clusterings (i.e., random ensembling would match it). This is less likely because the paper reports Totem beats ASW/VRC on average over multiple top trajectories, unlike Random.
Connectivity encodes differentiation βdirectionalityβ (not just branching topology). This is less plausible because connectivity is derived from undirected MST degree structure and the paper emphasizes selection/topology/milestones rather than proving directionality beyond Slingshot smoothing.
Explore by Goal
Quick Explanation
Long Explanation
Top Data Sources
Ask a Follow-Up
Key Insight
Analysis 
Hypothesis Graveyard
Potential Experiments
Discussion