- Trajectory inference and pseudotime analysis are important computational approaches in bioinformatics used to investigate how cells change their molecular characteristics during development, differentiation, activation, and disease progression. In biological systems, cells often move through a series of intermediate states rather than changing directly from one mature cell type into another. For example, stem cells may gradually differentiate into specialized cells, immune cells may transition from resting to activated states, and cancer cells may acquire molecular characteristics associated with invasion or treatment resistance. Trajectory inference attempts to reconstruct these cellular transitions from molecular data, while pseudotime provides an estimated relative position of each cell along an inferred biological progression.
- Traditional single-cell RNA sequencing (scRNA-seq) provides gene-expression profiles from thousands of individual cells at a particular sampling time. Although these measurements usually represent a snapshot rather than continuous observation, different cells within a tissue may occupy different stages of a biological process. By examining similarities and differences in their gene-expression profiles, computational methods can arrange cells along a possible developmental or functional trajectory. This allows researchers to investigate cellular transitions without necessarily observing every intermediate stage experimentally. However, an inferred trajectory is a computational reconstruction based on measured molecular similarities and should not automatically be interpreted as direct evidence of lineage relationships or actual cell movement.
- Trajectory inference generally begins with quality-controlled and normalized single-cell expression data. Highly variable genes are selected to capture meaningful biological differences, followed by dimensionality reduction using methods such as principal component analysis (PCA). A cell-neighbourhood graph is then constructed to represent relationships among cells. Clustering and graph-based approaches can identify major cellular populations and potential intermediate states. Trajectory algorithms use these relationships to estimate possible paths connecting different cellular states. Depending on the biological process, the inferred structure may be a simple linear progression, a branching trajectory, several independent paths, or a more complex network.
- Pseudotime is a numerical measure that represents the relative position of a cell along an inferred trajectory. It is called pseudotime because it does not usually measure actual chronological time. Instead, it orders cells according to their molecular similarity and their estimated progression from a selected starting state. Cells with low pseudotime values are generally positioned near the chosen starting point, while cells with higher values are located further along the inferred path. For example, in a stem-cell differentiation study, stem cells may receive early pseudotime values, intermediate progenitor cells may occupy middle positions, and differentiated cells may receive later values. The numerical scale depends on the method and trajectory, so pseudotime values should not be interpreted as hours, days, or other real-time units unless they have been independently calibrated against experimental time measurements.
- Trajectory inference methods differ in how they represent cellular relationships and calculate pseudotime. Monocle is one of the early and influential tools developed for ordering single cells along biological progressions. Monocle 2 uses reversed graph embedding to construct trajectories, while Monocle 3 supports graph-based trajectory analysis and principal graph learning. Slingshot uses cluster-based lineage inference and simultaneous principal curves to identify trajectories and estimate pseudotime. PAGA (Partition-based Graph Abstraction), implemented in Scanpy, summarizes relationships between cell clusters using a graph and can reveal potential branching structures. Other approaches include diffusion pseudotime (DPT), which uses diffusion maps to estimate progression through cellular states, and Palantir, which models differentiation potential and branching probabilities. These methods have different assumptions and may produce different results from the same dataset.
- The selection of a starting point, commonly called the root cell or root state, is particularly important in pseudotime analysis. A trajectory may be mathematically inferred in either direction, but biological interpretation requires an appropriate starting population. Researchers generally identify the root using known early-stage cell markers, experimentally defined progenitor populations, developmental time points, or prior biological knowledge. For example, a population expressing stem-cell-associated genes may be selected as the root of a differentiation trajectory. If the starting population is incorrectly specified, pseudotime ordering may be reversed or biologically misleading. Some methods can identify or estimate roots using additional information, but biological interpretation remains essential.
- Branching trajectories are important because many developmental and functional processes involve alternative cellular outcomes. A progenitor cell may differentiate into several specialized cell types, while an immune-cell population may develop into different functional states depending on environmental signals. Branching trajectory analysis attempts to identify where cells diverge into distinct paths and which genes are associated with each route. Monocle 3, Slingshot, PAGA, and other trajectory frameworks can support the investigation of branching patterns. Branch probabilities and lineage assignments can help describe alternative outcomes, although the presence of a branch in a computational model does not by itself prove that one cell population gives rise to another.
- Identifying genes associated with pseudotime is another important aspect of trajectory analysis. These genes may increase, decrease, or change their expression patterns as cells progress along a trajectory. Researchers can use statistical models to identify genes whose expression is associated with pseudotime and investigate their biological functions. For example, genes involved in stem-cell maintenance may decrease during differentiation, while genes associated with specialized cellular functions may increase. Methods such as tradeSeq use generalized additive models to identify genes that change along trajectories and compare expression patterns between lineages. Monocle also provides approaches for identifying genes associated with cellular progression. The resulting genes can be examined through Gene Ontology (GO), KEGG, Reactome, and other functional enrichment resources to identify biological processes and signalling pathways associated with cell-state transitions.
- RNA velocity provides complementary information about the possible direction of cellular transitions. It uses the relative abundance of unspliced and spliced messenger RNA to estimate short-term changes in gene expression. Unspliced RNA generally represents newly transcribed transcripts, while spliced RNA represents more processed transcripts. By modelling their relationship, methods such as velocyto and scVelo estimate the likely future direction of gene-expression changes for individual cells. RNA velocity can be combined with trajectory inference and pseudotime to improve the interpretation of cellular progression. However, velocity estimates depend on assumptions about RNA processing, transcriptional dynamics, and gene expression, and they do not directly measure cell lineage or guarantee long-term developmental outcomes.
- Trajectory inference can also be combined with lineage tracing and experimental time-course data. Lineage tracing uses genetic barcodes, cellular labels, or other experimental methods to follow the descendants of individual cells. Time-course experiments collect samples at multiple known time points, providing additional information about the chronological progression of biological processes. Combining these approaches with single-cell transcriptomics can help distinguish inferred cellular relationships from experimentally supported lineage relationships. Spatial transcriptomics can add anatomical information by showing where cells and intermediate states are located within tissues. Together, these technologies provide a more comprehensive understanding of cellular development and tissue organization.
- Several computational tools and programming frameworks support trajectory inference and pseudotime analysis. Monocle 3 provides trajectory learning, cell ordering, and visualization functions. Slingshot is commonly used with Bioconductor single-cell workflows, while Scanpy supports PAGA and diffusion pseudotime analysis. Palantir estimates differentiation potential and fate probabilities, and tradeSeq supports statistical analysis of gene-expression changes along trajectories. scVelo and velocyto provide RNA velocity analysis. Seurat can be used for preprocessing, clustering, dimensionality reduction, and integration before trajectory analysis. These tools are commonly used with R and Python, and results can be visualized using UMAP, principal component plots, trajectory graphs, heatmaps, and gene-expression curves.
- A typical computational workflow begins with scRNA-seq quality control, normalization, highly variable gene selection, and removal or assessment of low-quality cells and doublets. Researchers then perform dimensionality reduction and clustering to identify major cell populations. A trajectory method is selected according to the expected biological structure, and the root population is defined using experimental evidence or established markers. Pseudotime values are calculated, potential branches and lineage relationships are examined, and genes associated with progression are identified. Functional enrichment analysis is then used to interpret the molecular changes. Important results should be evaluated across biological replicates and, where possible, compared with time-course data, lineage tracing, RNA velocity, spatial information, or independent experimental studies.
- Trajectory inference and pseudotime have broad applications in developmental biology, stem-cell research, immunology, neuroscience, cancer research, regenerative medicine, and disease studies. In embryonic development, they can help reconstruct the differentiation of progenitor cells into specialized tissues. In immunology, they can investigate the progression of immune-cell activation, maturation, and exhaustion. In cancer research, they can reveal possible transitions between tumour-cell states associated with proliferation, invasion, metastasis, and treatment resistance. In neuroscience, they can help characterize neuronal and glial differentiation and changes associated with injury or disease. In regenerative medicine, trajectory analysis can support the study of stem-cell differentiation and tissue repair.
- These approaches are also valuable in animal genomics, veterinary medicine, and livestock research. Single-cell studies in cattle, sheep, goats, pigs, poultry, and other species can investigate developmental and functional transitions in muscle, mammary gland, reproductive organs, liver, blood, skin, and immune tissues. In skeletal muscle, trajectory analysis can examine the progression of muscle satellite cells from quiescent states through activation, proliferation, and differentiation. In mammary tissue, it can investigate changes in epithelial and supporting cell populations during development, lactation, and involution. In reproductive biology, it can help study germ-cell development, follicular maturation, and sperm-cell differentiation. In livestock immune tissues, it can reveal possible transitions associated with infection, inflammation, vaccination, and immune activation. Integrating these findings with quantitative genetics, genome-wide association studies, genomic selection, and gene regulatory network analysis may help connect cellular development and functional transitions with traits such as growth, milk production, fertility, disease resistance, and environmental adaptation.
- Despite their usefulness, trajectory inference and pseudotime analysis have important limitations. Different algorithms may produce different trajectories because they use different assumptions, distance measures, graph structures, and dimensionality reduction methods. Sampling may fail to capture rare or short-lived intermediate states, while batch effects, cell-cycle variation, stress responses, and technical noise can create misleading patterns. A continuous change in gene expression does not necessarily indicate a true developmental lineage, and two cells with similar expression profiles may have different biological histories. Pseudotime also depends on the selected root and the cells included in the analysis. Therefore, inferred trajectories should be treated as hypotheses that require biological interpretation and, where possible, experimental validation.
- Overall, trajectory inference and pseudotime analysis provide powerful approaches for studying cellular progression and dynamic biological processes using single-cell data. Trajectory inference reconstructs potential paths between cellular states, while pseudotime estimates the relative position of individual cells along those paths. When combined with gene-expression analysis, RNA velocity, lineage tracing, time-course experiments, spatial transcriptomics, and functional validation, these methods help reveal developmental pathways, cellular differentiation, immune responses, disease-associated transitions, and important biological processes in animals and other organisms.