Kinases are central regulators of protein function through phosphorylation, and their dysregulation contributes to many human diseases. Direct assays of kinase activity are often challenging or infeasible at scale, so researchers commonly infer kinase activity indirectly from measurements of substrate phosphorylation in phosphoproteomic data. However, many traditional inference methods simplify the underlying signaling architecture by treating kinase–substrate relationships as one-to-one or by otherwise ignoring that single substrates can be phosphorylated by multiple kinases. This simplification can obscure true signaling changes and limit accuracy, particularly when available data are limited.
The work summarized here motivates a more realistic modeling strategy that accommodates the directed, many-to-many structure of kinase–substrate networks. Modeling this complexity should improve both the fidelity of inferred kinase activities and the power to detect dysregulated signaling in disease contexts.
The authors introduce LIKA, a likelihood-based framework for inferring kinase activity from phosphoproteomic measurements. LIKA explicitly models directed graphs in which kinases are sources and substrates are targets, permitting multiple kinases to influence a single substrate. This network-aware statistical formulation uses a likelihood approach to estimate kinase activity parameters from observed phosphorylation patterns while accounting for the many-to-many mapping between kinases and substrates.
Key conceptual features of LIKA reported in the source include:
By retaining directed graph structure and modeling the joint contributions of multiple kinases to substrate phosphorylation, LIKA aims to produce more accurate and robust activity estimates than methods that do not consider these dependencies.
The authors report that LIKA was evaluated using both simulated data and empirical phosphoproteomic data from cell line experiments. According to the source summary, these analyses confirm LIKA’s robustness and accuracy in inferring kinase activity and in handling the complexities of many-to-many kinase–substrate networks. The source does not provide further numeric performance metrics, specific simulation settings, or detailed comparisons to named alternative methods in the summary; those details are available in the full preprint and associated materials.
These validation steps indicate that LIKA performs well across controlled simulations and real experimental data, supporting its use for downstream applications in biological and disease-focused studies.
As an application, the authors applied LIKA to a phosphoproteomic dataset derived from subjects with schizophrenia and matched controls. Using the network-aware inference provided by LIKA, the analysis identified novel dysregulated kinases and revealed altered patterns in protein phosphorylation networks associated with schizophrenia in the studied dataset.
The source summary does not enumerate which kinases were implicated, the number of dysregulated kinases found, or the effect sizes; those specifics are reported in the full article. The preprint status is emphasized: the findings have not been peer reviewed and should be interpreted accordingly.
The application demonstrates how a directed-graph, likelihood-based approach can uncover signaling alterations in a neuropsychiatric disorder where kinase signaling may play a role.
The authors provide implementation code and point to publicly available data to enable reproduction and further use of LIKA. The GitHub repository is cited in the source: https://github.com/lujingz/LIKA.
Funding sources disclosed in the preprint include the Simons Foundation and National Institutes of Health grants listed in the article metadata. The authors declared no competing interests.
Because this report is a preprint on bioRxiv, the analysis and conclusions have not been certified by peer review. Readers seeking detailed methods, numerical results, and lists of dysregulated kinases in schizophrenia should consult the full preprint and the supplementary materials linked from the article page.
Overall, LIKA represents a methodical step toward incorporating realistic directed network structure into statistical inference of kinase activity from phosphoproteomic data, with demonstrated utility in simulation, cell lines, and in identifying altered phosphorylation signaling in schizophrenia.