PMC

PERCEPTRON software architecture. PERCEPTRON web architecture is based on Model View Controller (MVC). Users can access this web service by signing up or logging in as a guest via PERCEPTRON graphical user interface (GUI) which is compatible with multiple web browsers. Secured PERCEPTRON web application program interface (API) acquires data from the front-end including (i) user credentials and (ii) parameters for protein search query, using https protocol. This information is stored in the PERCEPTRON database (DB). PERCEPTRON Core, which consists of the algorithmic pipeline, executes search using data stored in PERCEPTRON DB and PERCEPTRON Protein Repository. Processed results for each job are then moved to PERCEPTRON DB where they are stored. API returns these results to the front-end from where users can access them using PERCEPTRON GUI.

Cell-free perceptron enabling development of classifiers. a A perceptron scheme showing the inputs and their associated weights, the computation core, and the output. The perceptron computes the weights and actuates the weighted sum through an activation function. b Metabolic perceptron integrating multiple inputs and actuating an output. The benzoate actuator acts as the activation function of the perceptron reporting the sum of benzoate produced by the metabolic perceptron. Hippurate, cocaine, benzamide, and biphenyl-2,3-diol are the inputs of the metabolic perceptron fixed to 100 µM. The weights of the perceptron are the concentration of the enzymes calculated using the model made on weighted metabolic circuits (red circles). These weights are calculated to develop two classifiers using the metabolic perceptron and benzoate actuator. Full OR classifier c, [cocaine (C) AND hippurate (H)] OR benzamide (B) OR biphenyl-2,3-diol (F) classifier (d) are the two classifiers built using this metabolic perceptron. The Full OR classifier (c) classifies to OFF when none of the inputs is present and it passes an arbitrary threshold to ON when any of the inputs or their combinations are present. The second classifier (d) performs a more complex computation. The shading represents the arbitrary threshold that allows for perceptron decision making and the panel of OFF and ON at the top of the bars are the expected output of the classifiers. All data are the mean and the error bars are the standard deviation of normalized values from three measurements and red circles are the model predictions (gray dots on each bar are individual values from which the mean and SD were calculated). (RFU relative fluorescence unit). Source data are provided in the Source Data file

(a). Multilayer Perceptron error (X: Availability vs. Y: Predicted Ranking). (b). Multilayer Perceptron error (X: Security vs. Y: Predicted Ranking). (c). Multilayer Perceptron error (X: Reliability vs. Y: Predicted Ranking). (d). Multilayer Perceptron error (X: Cost vs. Y: Predicted Ranking). (e). Multilayer Perceptron error (X: Ranking vs. Y: Predicted Ranking).

Exact perceptron outputs that represent sigmoid of the weighted sum of the inputs and the molecular perceptron outputs that compute sigmoid of the weighted sum scaled by a factor 1/32 for the 100 input vectors for: (a) Perceptron A, (b) Perceptron B, (c) Perceptron C. The x axis corresponds to input vector number.