Svetlov, D. | Search bioRxiv

bioRxiv2026

The Stochastic System Identification Toolkit (SSIT) to model, fit, predict, and design experiments

Biological data is prone to both intrinsic and extrinsic noise and variability between experimental replicas. That same stochasticity and heterogeneity can carry information about underlying biochemical mechanisms but, if not incorporated in modeling and probabilistic inference, can also bias parameter estimates and misguide predictions and, subsequently, experiment design. Mechanistic inference typically requires lengthy simulations (e.g., the Stochastic Simulation Algorithm (SSA)); approximations to chemical master equation (CME) solutions that lack rigorous error tracking; or deterministic averaging that lacks the complexity necessary to reflect the data. We introduce the Stochastic System Identification Toolkit (SSIT) - a fast, flexible, and open-source software package available on GitHub that makes use of MATLABs efficient and diverse computational architecture. The SSIT is designed for building, simulating, and solving chemical reaction models using ODEs, moments, SSA, Finite State Projection truncations of the CME, or hybrid methods; sensitivity analysis and Fisher information quantification; parameter fitting using likelihood- or Bayesian-based methods; handling of experimental noise and measurement errors using probabilistic distortion operators; and sequential experiment design that empowers users to save time and resources while gaining the most information possible out of their data. The SSIT also offers advanced modeling tools, including model reduction methods for increased efficiency and joint fitting of models and datasets with overlapping reactions or parameters. To facilitate the ease and speed of use, the SSIT provides a graphical user interface and ready-made, adaptable pipelines that can be run in the background from commandline or high-performance computing clusters. We demonstrate features of the SSIT on two experimental datasets: the first consists of published mRNA count data that reflect Saccharomyces cerevisiae yeast cell response to osmotic shock using single-cell single-molecule fluorescence in situ hybridization; the second consists of single-cell RNA sequencing measurements of 151 activating genes in breast cancer cells following treatment with dexamethasone. Author summaryWe present the Stochastic System Identification Toolkit (SSIT) to model, fit, and predict any data that can be interpreted as changing populations or counts through time, including but not limited to single-cell experiments, economics, epidemiology, ecology, sociology, agriculture, and biotechnology. The SSIT was constructed particularly for stochastic modeling, which is important for systems whose states may experience significant fluctuations from mean behavior, thus affecting the inference of the underlying rate parameters and predictions of subsequent behavior. The SSIT provides statistical inference tools for parameter estimation; sensitivity analysis and information calculation; handling of distortions to probability distributions caused by experimental or measurement processes (e.g., dropout in single-cell RNA sequence data and total fluorescence intensities versus spot counting/puncta analysis); and quantitative design of experiments. The SSIT also offers a variety of complex modeling tools, including model reduction methods and fitting of combined models/datasets that share some behavior but remain distinct (e.g., different genes responding a single stimulus). The SSIT generates pipelines for easy, efficient analyses to run in the MATLAB environment, in the background on commandline, or on high-performance computing clusters, thus facilitating users to make informed, time- and cost-effective decisions about their next set of experiments.

bioinformatics↗

bioRxiv2024

Sequential design of single-cell experiments to identify discrete stochastic models for gene expression.

Control of gene regulation requires quantitatively accurate predictions of heterogeneous cellular responses. When inferred from single-cell experiments, discrete stochastic models can enable such predictions, but such experiments are highly adjustable, allowing for almost infinitely many potential designs (e.g., at different induction levels, for different measurement times, or considering different observed biological species). Not all experiments are equally informative, experiments are time-consuming or expensive to perform, and research begins with limited prior information with which to construct models. To address these concerns, we developed a sequential experiment design strategy that starts with simple preliminary experiments and then integrates chemical master equations to compute the likelihood of single-cell data, a Bayesian inference procedure to sample posterior parameter distributions, and a finite state projection based Fisher information matrix to estimate the expected information for different designs for subsequent experiments. Using simulated then real single-cell data, we determined practical working principles to reduce the overall number of experiments needed to achieve predictive, quantitative understanding of single-cell responses.

systems biology↗

bioRxiv2021

NMPylation and de-NMPylation of SARS-CoV-2 Nsp9 by the NiRAN domain

Nsp12, the catalytic subunit of SARS-CoV-2 RNA-dependent RNA polymerase (RdRp), contains two active sites that catalyze nucleotidyl-monophosphate (NMP) transfer (NMPylation). RNA synthesis is mediated by the RdRp active site that is conserved among all RNA viruses and has been a focus of mechanistic studies and drug discovery. The second active site resides in a Nidovirus RdRp-Associated Nucleotidyl transferase (NiRAN) domain. Both catalytic reactions are essential for viral replication, but the mechanism and targets of NiRAN are poorly characterized. One recent study showed that NiRAN transfers NMP to the first residue of RNA-binding protein Nsp9. Another study reported a structure of SARS-CoV-2 replicase with an extended Nsp9 in the NiRAN active site but observed NMP transfer to RNA instead. We show that SARS-CoV-2 Nsp12 efficiently and reversibly NMPylates the native but not the extended Nsp9. Substitutions of the invariant NiRAN residues abolish NMPylation, whereas a substitution of a catalytic RdRp Asp residue does not. NMPylation is inhibited by nucleotide analogs, pyrophosphate, and bisphosphonates, suggesting a path for rational design of NiRAN inhibitors. We hypothesize that Nsp9 remodels both active sites of Nsp12 to support initiation of RNA synthesis by RdRp and subsequent capping of the product RNA by the NiRAN domain.

biochemistry↗