MLPERCEP: Multiple layer perceptron implementation for the biological problems.
To use this software, you have to download and install the NET framework and the software. The links are given below.
Download NET framework from windows
Multiple layer perceptron also know as Feed Forward Back-Propagation algorithm is a class of supervised neural Network algorithm, and has lot of application in solving many of the complex mathematical problems.
The algorithm has an important role wherever we have two or more groups and we intend to develop a classifier to distinguish those groups. The flowchart would be
1) Provide the network with sufficient examples
2) Let the network learn by changing its weight
3) Test the efficacy on the fresh datasets.
The general architecture of the algorithm is shown below.
It has an input neuron layer, a hidden neuron layer and an output neuron layer. The number of input neuron will depend on the number of inputs. The inputs can be binary or floating numbers. However the results are more accurate if values within -1 and 1 are used. Currently the software supports only binary outputs i.e. 0 or 1 or any value between them. The hidden layer is the generalization layer and can be varied by the user. The very high number of hidden neurons may lead to quicker learning, but may fail to generalize. However very less number of neurons may not allow the network to train at all. Some of the other fine tuning parameters provided to the user are learning rate, momentum, error cut-off, maximum number of iterations and leave one out cross validation.
The format of input data for the training and testing module is
Rows 5
Cols 4
Sample1 0 0.2 0.6 0.1 0.9
Sample2 1 0.8 0.4 0.9 0.7
Sample3 0 0.1 0.45 0.3 0.96
Sample 4 1 0.9 0.7 1 0.99
Sample5 1 0.8 0.77 0.8 0.8The rows indicates the number of inputs. The column indicates the number of input parameters. The second column is the expected output. The parameters in the training file can be space or tab delimited. Testing data should be in a similar way. However there is no need for the second column i.e. expected result More details.
Download sample Dataset: Training data Testing data
There are special formatting modules provided for the gene expression and sequence dataset.
Design:
The MLPERCEP software is a collection of individual programs written in C language. Each of the C programs is complemented with the graphics user interface written in C# and the individual sets can be accessed from the main user interface also in C#. The C programs extensively use dynamically memory allocation and dynamic utilization of hard disk space to practically make the program handle extremely large networks and datasets. To run the software, one needs to install the .NET runtime environment from windows.
If you have any questions or problems contact jpmehta@bioinformatics.org
You can also try using the old version. It is faster than the current version, but does not include run-time randomization of samples and equal loading of positive and negative examples.
