Details of each function
Data Access
You can easily import your existing data, such as table data (CSV format, can mix numbers and strings) and chemical structure data (SMILES format).
Experimental conditions and chemical structure generation
Candidate experimental conditions and chemical structures can be automatically generated in the thousands or even tens of thousands.
Experimental Design
You can statistically narrow down a large number of experimental conditions to effective candidates for initial experiments.
Descriptor Calculation
Chemical structures are converted into numerical data. Both homopolymers and copolymers can be handled.
Data Visualization
Various statistical graphs and dimensionality reduction make it easy to understand the distribution of features and the characteristics between features.
Data Preprocessing
Features can be converted, added, or deleted to adjust them to a form suitable for model building.
Regression analysis
You can consider building a model with high prediction accuracy from various algorithms. In addition, the regression coefficients and importance of each feature are listed, allowing you to check the basis for prediction.
Adaptive Experimental Design
Based on existing experimental data, potential candidates for further experimentation are suggested.
Bayesian Optimization
In adaptive experimental design, this is an effective method when the target value is far from the training data.
Time Series Data Analysis
Models can be built using time-varying data, and predictions can be made at any time based on data measured in real time, enabling soft sensors and anomaly detection to be put into practical use.
Model Optimization
It is easy to automatically optimize parameters and compare the predictive accuracy of various models.
Inverse analysis
After building a predictive model for physical properties and activity, you can determine the conditions for achieving the target values. You can also perform direct reverse analysis, which is not available with other services.
Missing value imputation
When there are missing parts in the training data, the missing values are automatically estimated from the information of the entire feature set.
Classification
Category classification predictions can be made using qualitative data as the objective variable.
mixture calculation
Combining raw material composition data with the physical properties and characteristics data of the raw material itself, new feature quantities are calculated using various calculation methods.
Spectrum
Numerical data from spectra and profiles from analytical instruments can be processed to predict composition and physical properties.
Reagent Database
Data on candidate reagents is extracted from over 490,000 compounds from Fujifilm Wako Pure Chemical Corporation, and after predicting their physical properties and characteristics, you can directly access the product information site for the optimal reagent.