Integrated artificial intelligence boosts the power of this chemical database for three-dimensional molecular structures
ChemBrain is a searchable chemical database for three-dimensional molecular structures with integrated artificial intelligence.
ChemBrain software's integrated artificial intelligence makes ChemBrain capable of learning, says Silverdale.
Invaluable in the prediction of any molecular properties as well as for use as laboratory notebook.
Extremely simple and user-friendly input of three-dimensional molecules, supported by a fast geometry-optimiser, it stores any kind of data for use in artificial neural network calculations.
ChemBrain enables graphical input of molecules in an immediate three-dimensional form.
Generation of new molecules is extremely simplified, in that the user may either start with pre-generated fragments or use any molecule from the ChemBrain database and either modify them or combine them with any other fragment, which latter may even be immediately generated on screen by duplicating segments or the entire molecule already drawn.
At any time the structure can be optimised starting a fast force-field geometry optimiser.
The ChemBrain data sheet window shows all the data of a molecule that are stored in the ChemBrain database.
Addition of any further properties as well as any kind of notes is enabled in this window.
The three-dimensional molecular structure can be rotated freely.
A simple double-click on the graphical screen opens the graphical input window using this molecular structure as a starting point for a new molecule entry.
The radial distribution function, which is used for similarity search and neural network calculation purposes, can be optionally based on various parameters such as atomic mass, net charge or ability to be polarised.
Prediction of properties is one of the particular strengths of ChemBrain.
The user has the option of selecting between several architectures of neural networks: mapping, modelling or classification.
The program then searches its database for the most appropriate pre-calculated neural network and uses it for the prediction of the requested property or, if none is found, suggests training a new neural network based on molecules in the ChemBrain database structurally most similar to the query molecule.
Depending on the selection of the architecture of the neural network, ChemBrain generates a prediction window showing on the left side the query molecule and the corresponding prediction value of the queried property, and on the right side the underlying neural network, either as a map or as a list, showing the molecules used for training.
The rectangle framed in red indicates the position of the query molecule in the neural network in relation to the ones used for training.
Another ChemBrain strength is its potential to use its database for the association of the three-dimensional molecular structures with any of the stored molecular properties by applying various architectures of neural networks.
ChemBrain provides several architectures of neural networks as tools for the prediction of properties.
By training the neural network with a series of steroids and their biological activity parameter using a back-propagation algorithm for classification purposes, the neural network is stored in a specific way for potential use in the prediction of this activity for as yet unknown similar compounds.
Besides its learning and prediction capabilities, ChemBrain provides all the tools that can be expected from a chemical database.
ChemBrain uses a specific fragment search algorithm for this purpose which generates and compares fragments of up to (at present) 12 atoms, thus ensuring that only hits are found that really contain the queried fragments.
An interesting property of ChemBrain is its capability of comparing the 3D structure of a molecule with others based on various atomic properties, a feature that is inherently used in ChemBrain's neural network calculations.
Searches may be made using atomic mass, alternatively, the atomic net charge, the atomic polarisability or others may be applied.
ChemBrain also allows the user to restrict this kind of search to molecules of a certain class or with a known property One of the main advantages of ChemBrain over ordinary chemical databases is its potential to apply artificial intelligence on its own database.
In order to take advantage of this feature, the user need not be familiar with the principles of neural networks, (although its is advantageous; the online help provides some useful introductory explanations) since the program guides him in a self-explanatory way.
After defining the task, the program suggests the network strategies available for the task and allows the user to optionally select various further data.
A more experienced user may also modify the RDF parameters and/or the standard setup of the neural network for which the corresponding forms are then opened.
If these latter features are not selected, ChemBrain will use the best parameter set for the task.
