Review papers with raw data transparency

What the authors did

The authors performed a PRISMA-guided systematic review (search June 15 2024) that included real or pseudo longitudinal multi-omics studies and identified 143 eligible studies (125 applied, 18 methods), cataloging data types, sampling regimes, and computational approaches across host and microbiome contextsPRISMA search and study counts

Key findings (evidence-cited)

• Most studies are exploratory; only a minority employ integrative longitudinal frameworks designed for time series (MOFA/MEFISTO, DBNs, PALM, SNF, iCluster) rather than sequential single-omics analysesIntegrative frameworks versus exploratory analyses
• Data sparsity, nonmatching modalities across time points, pseudo time-series and uneven sampling are recurrent practical constraints that shape method choice and interpretationData sparsity and pseudo time-series
• Adaptive, probabilistic and ML methods (Gaussian processes, LMMs, DBNs, RNN/LSTM, transformers) are promising but adoption is limited by computational cost, interpretability and availability of maintained implementationsAdaptive and ML methods limitations
• There is an adoption gap: many methodological tools exist but code, documentation, maintenance and ease of use are inconsistent — limiting reproducibility and reuseAdoption gap description

Strengths of the review

• Systematic PRISMA approach, explicit inclusion of pseudo time-series to capture destructive sampling designsPRISMA and pseudo-time inclusion
• Practical, method-focused evaluation criteria (performance, interpretability, ease of use, maintenance) that matter for real-world adoptionMethod evaluation criteria

Limitations and blindspots

1. Single-source dependence in this review can only summarize existing literature; it cannot quantify comparative method performance because few method papers reported standardized benchmarks or shared data (limiting objective comparison). The authors acknowledge that many method repositories were deprecated or not maintainedMethod code maintenance problems
2. The review is descriptive and qualitative; it does not present registered reanalyses or meta-benchmarks (so claims about relative performance rest on primary authors summaries rather than standardized re-evaluation). This reduces the strength of any ranking claims.
3. Because of heterogeneous reporting across studies (sampling schedules, preprocessing, normalization choices), cross-study generalization is limited — the review flags this heterogeneity but cannot fully control for itHeterogeneous reporting and sparse data

Specific, actionable technical recommendations

• Encourage standardized benchmark datasets and challenge tasks (public longitudinal multi-omics sets with held-out test splits) so methods can be compared reproducibly.
• Promote containerized, versioned software with long-term maintenance plans and clear tutorials; flagging of deprecated repos is vital and already observed by the authorsNeed for maintained code and docs
• Adopt hybrid modelling: use mechanistic models (ODEs, agent-based) to encode domain constraints plus ML for residual structure; the review highlights mechanistic modeling as underused but promisingMechanistic models recommendation
• Prioritize robust missing-data handling and compositional-aware methods (microbiome proportions require specific transforms) — the review notes inconsistent addressing of compositionality and transformations.

Quantitative critical scores (expert judgement)

See below for short explanations of each score.

Why these scores

• Novelty 6 — first comprehensive systematic mapping focused on longitudinal multi-omics in microbiome research, but conceptual components (MOFA, DBN, LMMs, RNNs) are established.
• Quality 7 — solid PRISMA workflow, careful annotation, dual review per study; limitations are descriptive nature and absence of standardized re-benchmarks.
• Generality 7 — covers multiple host and environment systems and many methods; heterogeneity limits universal prescriptive claims.
• Usefulness 8 — practical for method selection, highlights adoption barriers and priority directions for field; valuable to practitioners and method developers.
• Reproducibility 6 — authors provide code and tables (claimed), but many primary-method repositories are unmaintained and many reviewed studies lack harmonized data sharing (reducing reproducibility across the field)Code and reproducibility note
• Explanatory depth 6 — good survey-level synthesis and identification of methodological categories, but intrinsic to reviews: mechanistic causal depth and new theory are limited.

Novel hypothesis suggested (testable)

1. Hybrid mechanistic-deep models that constrain learned dynamics by mass-balance and known metabolic reactions will outperform black box temporal ML in identifying causal microbe–metabolite–host axes in longitudinal multi-omics when missing data exceed 30% (test via held-out prediction and causal mediation on curated longitudinal datasets).

How to improve future versions of this review

• Include standardized benchmark re-analyses of a small number of representative longitudinal multi-omics datasets (public) to quantify performance across methods rather than rely solely on author reports.
• Provide a machine-readable registry (metadata schema) of each reviewed method with installation status, container, last commit date, license and demo dataset for rapid community vetting.

Bottom-line (evidence-linked)

The review is a timely, well-executed mapping of methods and practices in longitudinal multi-omics microbiome research that usefully identifies major practical bottlenecks (sparse unmatched modalities, pseudo time-series, adoption gap for integrative tools) and points to promising directions (mechanistic models, adaptive ML, standardized benchmarks). These conclusions and recommendations are explicitly stated in the article and are supported by the authors textual syntheses and the counted study corpusSummary and recommendations excerpt

Practical next steps for researchers

1. Publish containerized code and example longitudinal datasets with clear evaluation metrics.
2. Run cross-method benchmarks using unified preprocessing (compositional transforms for microbiome data) and missing-data regimes to reveal strengths/weaknesses.
3. Fund and maintain an open repository of validated longitudinal multi-omics analysis pipelines (community governance).