Datachemical LAB
AI/Machine Learning Cloud Service for Experimental/Manufacturing data analysis
What is Datachemical LAB?
Datachemical LAB is a SaaS product that can handle data analysis and machine learning programs useful in the chemical and engineering fields with simple operations.
A systematic program from molecular / material design to manufacturing process design / control is fully equipped on one platform, and customers who are engineers can execute advanced data analysis / prediction model construction by themselves without programming. Customers can reduce costs and improve development efficiency on various chemical and engineering development themes, and can acquire new knowledge through data science.
We now have Japanese version only. Will release English version soon.
Usage example
It can be used in the development of all kinds of materials such as inorganic, organic, polymer materials, and foodstuffs.
From the experimental data of 20 to 30 samples, experimental conditions that are highly likely to achieve target performance are predicted.
From automatically generated ten thousands of experiment candidates, a condition to be tested first is statistically select .
From quantified and learned chemical structures data, a chemical structure that reaches target physical properties is estimated.
Manufacturing processes are managed by estimating factors, which real-time measure is difficult, from sensor data of a chemical plant.
Application
Chemoinformatics
(Molecular design)
Will quantify chemical structures and perform machine learning between the structures and their properties / activity, and construct a model that predicts properties, activity and chemical reactions from new chemical structures before synthesis. In addition, it is possible to estimate chemical structures and their reaction pathways that achieve target physical properties and activity.
Materials informatics
(Material design)
Will learn experimental data and construct a model that predicts properties and activity of materials from new experiments and manufacturing conditions. In addition, it is possible to estimate experimental and manufacturing conditions to obtain target materials from raw materials.
Process informatics
(Process design and control)
Will design equipment and processes to synthesize target materials using process data. In addition, can build and operate a system called a soft sensor that estimates factors that are difficult to measure in real time from plant sensor data at any time, and a model that detects abnormal plant conditions.
Functions
Data Access
Can easily input your own experimental data such as table data (CSV format, which can be mixed with numerical values and character strings) and chemical structure data (SMILES format).
Experimental Conditions / Chemical Structures Generation
Can automatically generate candidates for experimental conditions and chemical structures in thousands or tens of thousands of units.
Design of Experiments
Can statistically select effective candidates for the first experiment from a large number of experimental conditions.
Descriptor Calculation
Can convert chemical structures into numerical data. Can also handle homopolymers and copolymers.
Data Visualization
Can easily grasp distributions of features and relationships between features by using various statistical graphs and lower dimension models.
Data Preprocessing
Can convert, add, and delete features and make data suitable for model constructions.
Regression Analysis
Can construct a model with high prediction accuracy from various algorithms. In addition, can confirm the basis of the prediction from lists of regression coefficients and importance of each feature.
Adaptive Design of Experiments
Based on existing experimental data, can propose effective candidates for the next experiment.
Bayesian Optimization
In adaptive design of experiments, it is an effective method when target values are far from training data.
Time Series Data Analysis
Can construct a model using time-varying data and make predictions from real-time measured data at any time.
Model Optimization
It is easy to automatically optimize parameters and compare prediction accuracy of various models.
Inverse Analysis
After constructing a predictive model of properties and activity, can determine conditions to achieve target value.
Missing Value Imputation
When there is a missing part in training data, the missing value is automatically estimated from the information of the entire feature quantity.
Classification
Categorical prediction can be performed using qualitative data as the objective variable.
Mixture Calculation
New feature values are calculated by various methods by multiplying raw material composition data and physical property data of the raw material itself.
Features
Seamless use
Can deal with areas of molecule, material, and process design that have been considered individually on one platform, and can improve total efficiency from experiments in the laboratory to mass production.
Prediction accuracy
By statistically selecting experimental candidates for initial data collection and automatically optimizing from more than 20 algorithms based on the obtained data, can obtain high prediction accuracy from a small number of experiments and can achieve a development goal in a short period of time.
Ease of use
Can execute on a simple UI without programming, preventing analysis mistakes that data science beginners tend to make. With a wealth of manuals, can easily get started without advanced data science skills.
Process prediction with desktop App
Based on results of optimization studies in cloud, can conduct soft sensors and abnormal conditions detection on new manufacturing process data by App. It works offline, so if you install it on your PC, you can use it without being affected by network conditions at a manufacturing site.
Security
Plan
Corporate plan
Various companies can utilize on a variety of technical themes from development of next-generation materials to improvement of existing products.
We will correspond your individual inquiries by e-mail, web interview, etc. We will provide a separate consulting service for difficult themes that Datachemical LAB cannot handle.
To maintain the quality of service, we will start with a limited number of 100 users across the Datachemical LAB. Therefore, even if you contact us, we may make you wait for service, but we appreciate your understanding. We will increase the capacity.
Academic plan
We offer Datachemical LAB at special price to users of research and educational institutions.
We will support technical inquiries via group chat.
Both plans offer free trials.
Strengths of operating company,
Datachemical, Inc.
Based on the knowledge of Chemoinformatics, Materials Informatics, and Process Informatics, of Associate Professor Hiromasa Kaneko, Faculty of Science and Technology, Meiji University, Japan, we offer data science business specializing in the chemical and engineering fields.
Accumulating the latest research results of data analysis and machine learning utilization in the chemical and engineering fields, we contribute to solving the technical problems of a wide range of our customers by updating Datachemical LAB platform and consulting on advanced technical projects.