CFS_BHDS_AS Software Basic Info
CFS_BHDS_AS Software WIKI Documentation
Software Name: CFS_BHDS_AS (Curve Fitting Software Based on High-Dimensional Search and Azimuth Statistics)
Version: v1.0
Developer: Shengxuan BIAN
Contact Email: shixiu_yakuchi0324@qq.com
Document Date: 3 October 2025
1. Summary and overview
Software Overview
CFS_BHDS_AS is an intelligent curve fitting software based on high-dimensional space search and orientation statistics, which is specially used to optimize the parameters in biological system dynamics models.
It supports two core application scenarios:
- Import simple and easy-to-obtain wet experimental data to predict difficult-to-measure data.
- Take ideal wet test data as the target and data to be improved as the starting point; by fitting the behavior of observed parameters, find the key parts that help the wet test achieve the ideal state.
Core functions
- Automatic parameter optimization: Fit experimental data with the prediction of the ODE model.
- Real-time visualization: Dynamically display the fitting process and parameter change behavior.
- Robust search algorithm: Combine reinforcement learning with orthogonal search strategy to effectively escape from local optimal solutions and avoid wasting computing power.
Target User
Computational biology researchers, synthetic biology researchers, and researchers who need to deal with ODE model parameter fitting.
2. Quick Start
2.1 system requirements
- Software dependencies: R (>=4.5.0), RStudio (>=2025.05.0)
- Hardware dependencies: Recommended configuration (e.g., Intel i7 or higher performance CPU)
- Installation and startup (step-by-step guide)
- Get the software package.
- Open the
CFS_BHDS_AS.Rprojproject file in RStudio. - Run the load command on the console: load("CFS_BHDS_AS.Rdata").
- Input: CFS_BHDS_AS() and press CTRL + Enter to start running.
- View the unpacked file structure (Figure 1):
Figure 1-File after unpacking the software package
2.2 The first analysis: Five minutes to master
- Prepare your experimental data file (
Exp files/Exp.xlsx). - Prepare or select your ODE model file (
ODE files/). - Enter information as prompted by the software.
- View and understand the result file (
PBPP_RESULT/).
3. Detailed workflow
3.1 Pre-analysis: documentation
3.1.1 Experimental data file (Exp files folder)
- Format requirements: .xlsx
- Data volume suggestion: Time point > 48, recommended > 240
Figure 2-Exp.xlsx Content example
3.1.2 Model file (ODE file folder)
- ODE equation format requirements: Refer to the notes in the model file for modification guidelines.
- Each model file contains a modified guide in the form of notes (Figure 3).
Figure 3-Each model file contains a modified guide in the form of notes
3.2 Analysis in progress
Software configuration and operation: Refer to software prompts and README.txt for specific steps.
3.3 Analysis result management
- Default output directory:
PBPP_RESULT(the directory where the software reads and outputs the latest data). - Directory renaming rule: Rename the
PBPP_RESULTdirectory to any name to allow the software to automatically create a newPBPP_RESULTfolder for subsequent analysis.
4. Architecture and design principles
4.1 Methodology
The core methodology of CFS_BHDS_AS is based on high-dimensional space coordinate transformation, and the specific logic is as follows:
- For any Ordinary Differential Equation (ODE) model, all parameters α associated with its current prediction curve are defined as predictive coordinates A (each dimension corresponds to a specific parameter value).
- Experimental data curves β correspond to unique parameter combinations from the ODE model perspective; similarly, there is at least one target coordinate B with distinct parameter combination values (constituting its dimensional information).
- The Euclidean distance AB between A and B represents the degree of divergence between the prediction curve α and the experimental curve β.
- Coordinate adjustment strategy:
- Continuously adjust the predicted coordinate A using spatial vectors to approach the target coordinate B.
- If approaching efficiency decreases, find a new spatial vector perpendicular to the current trajectory direction (minimizing Euclidean distance AB) as the next coordinate prediction guidance.
- Distance calculation: Use R language function
base::outer(x, y, "-")to calculate the distance information matrix between α and β (after identical weighting for each coordinate), and then compute Mean Squared Error (MSE) to measure their difference.
- Dynamic vector adjustment (reinforcement learning + orthogonal constraint):
- Use reinforcement learning to record the historical performance of each parameter direction through a weighted environment, and probabilistically generate new movement vectors based on this data.
- When optimization efficiency declines, orthogonal constraints generate new vectors perpendicular to the current direction to maintain search momentum.
- Use a whitelist/blacklist mechanism to prevent redundant and ineffective searches, and perform local fine-tuning on high-performing directions.
Essentially, this transforms the problem of "how to adjust parameters to make the prediction curve more similar to the experimental curve" into "how to guide the distance between two points to decrease in high-dimensional space".
4.2 Software design
4.2.1 Modular design
Separate ODE files, Exp files, and result output modules for easy management and reuse.
4.2.2 User interaction design
Provide a graphical interface based on the R environment to reduce the threshold of command line operation.
4.2.3 Design trade-offs
- Universality and specificity: The software is suitable for a variety of ODE models, but its performance highly depends on the scientificity of the model structure and initial parameters provided by users.
- Accuracy and speed: The
BestTRDthreshold allows users to trade off fitting accuracy and computation time.
5. Key choices in software development
The software adopts two core design choices to ensure stability and effectiveness:
- Missing value detection mechanism: A large number of missing value detection codes are set up to prevent system crashes caused by incomplete data.
- Unreasonable parameter correction (reinforcement learning): Through the reinforcement learning system, unreasonable parameter adjustments made by users can be effectively transformed into reasonable parameters to ensure effective calculation.
6. Maintenance and support
For software maintenance, technical support, or problem feedback, please contact the developer via email:
Contact Email: shixiu_yakuchi0324@qq.com
Please describe the problem in detail (e.g., error information, operation steps, data format) when sending an email to improve the efficiency of problem solving.
7. Frequently Asked Questions (FAQ)
Q1: How to deal with "long duration display of temporary files"?
A1: This issue occurs when the input ODE file has computational errors (e.g., zero division, calculation of excessively large or small numbers), causing the software to be stuck. It is recommended to:
- Check the ODE equation logic for mathematical errors (e.g., denominator variables that may be zero).
- Verify the parameter range settings in the ODE file to avoid overflow/underflow.
Q2: Why does the fit fail or report an error?
A2: Please first check the following two core factors:
- Data-model matching: Verify the format of wet experiment data and ODE files. You must input data that matches the ODE file specifications. For example, if you input a wet experiment plasmid concentration curve but fail to correctly specify the second column name of
Exp.xlsx, the software will be unable to map data to model parameters. - Curve rationality: Although the software has some adaptability to noise curves, using incorrect curve data (e.g., randomly guessed curve shapes) will definitely cause fitting failure. The software can only learn from slightly "casual" curves (Figure 4), not completely irrational data.
Figure 4-The software can learn to have a slightly "casual" curve
Q3: How to replace the model without affecting previous analysis results?
A3: Follow the steps below:
- Terminate the running software.
- Rename the existing
PBPP_RESULTfolder to any name (it is recommended to add a data identifier, e.g.,PBPP_RESULT_20251003) to save previous results. - Modify the ODE file code (e.g., replace the ODE equation or adjust initial parameters).
- Restart the software; it will automatically create a new
PBPP_RESULTfolder to store the new analysis results.
8. Software limitations
- Model type limitation: The software can only accept ODE models at present, and does not support other types of dynamic models (e.g., partial differential equations, stochastic differential equations).
- Curve shape length limitation: There is a length limit to the shape of the detection curve; the software can only observe and fit data within the supported length range (the more data provided, the more accurate the fitting result).
- Language dependency: The software is not yet independent of the R language; it must run in the R/RStudio environment and cannot be deployed as a standalone executable program.
9. Deployment and integration
9.1 Standard deployment process
Refer to the "Quick Start" chapter for deployment steps, and focus on verifying the following two points to ensure normal operation:
- The R version is ≥4.5.0 and the RStudio version is ≥2025.05.0.
- The software package is completely unpacked, and the
CFS_BHDS_AS.Rprojfile can be normally opened in RStudio.
9.2 Visualization and reporting integration
The software's output results (including fitting curves, parameter change charts, etc.) are in standard image formats (e.g., PNG/JPG), which can be directly embedded into academic papers, project reports, or presentation slides without additional format conversion.
Key integration suggestions:
- When inserting results into papers, it is recommended to retain the original image resolution (≥300 DPI) to ensure clarity.
- When citing results in reports, mark the corresponding
PBPP_RESULTfolder name to facilitate result tracing.
EasyDock: An Automated Molecular Docking Platform Empowering Synthetic Biology Research
In the fields of synthetic biology and drug design, molecular docking is a crucial tool for understanding the interactions between biological macromolecules and small molecules. However, traditional molecular docking workflows often require complex software configuration, tedious parameter adjustment, and specialized computational biology expertise, creating a technical barrier for many iGEM teams. We previously attempted to perform rapid molecular docking using the existing Free_Cloud_Docking_2D_3D project on Google Colab, but this still required manual clicks, parameter adjustments, and dependency on known ligand positions to set the docking box—making it less suitable for proteins without pre-bound ligands. Additionally, we encountered issues such as network latency, poor performance, and high network dependency on cloud computing platforms. To address these challenges, we migrated the Free_Cloud_Docking_2D_3D project to local computers and developed EasyDock—a user-friendly, automated molecular docking platform—enabling every iGEM team to conduct professional molecular docking analyses on personal computers with ease. The following is a schematic diagram of the software usage and result output:
