Skip to main content

Cheminformatics

These programming exercises are designed to introduce developers to the Datagrok platform cheminformatics capabilities. The exercises are based on your knowledge obtained in exercises.

Table of contents​

Basic exercises in cheminformatics​

Exercise 1: Search for most common structures​

You will learn: How to employ functions from external packages in your own package.

Prerequisites: "Cheminformatics".

Statement of the problem. Write a function that reads a file containing SMILES, determines the associated maximal common substructure (MCS), and computes the mutual similarity scores for molecules and the MCS.

Input data. Files > App Data > Chem > sars_small.csv

Solution, step-by-step.

  1. Let's call our function findSimilarToMCS, we place its definition in ./src/package.ts inside our package. This function takes a single input — a dataframe df. For the sake of simplicity, we suppose that the column with SMILES is df.col('smiles'):

    //name: findSimilarToMCS
    //input: dataframe df
    export async function findSimilarToMCS(df: DG.DataFrame) : Promise<void> {
      ... // your code goes here
    }

  2. Employ the asynchronous function FindMCS from Chem package. Since we're calling a function from an external package, we should use grok.functions.call:

    const mcsValue = await grok.functions.call('Chem:FindMCS', {'smiles': 'smiles', 'df': dataframe, 'returnSmarts':
    false});

  3. Having obtained the string mcsValue, create a new column in df, whose cells are filled with this value:

    • Create an Array of the appropriate length, filled with mcsValue.
    • Feed this array to the constructor DG.Column.fromList() to get the desired mcsCol object.
    • Assign semantic type Molecule to the newly created column, with the help of col.semType(...). Similarly, associate Molecule cell renderer with the help of col.setTag(...) method.

  4. To compute similarity scores, we can call the getSimilarities() function of Chem package, which takes as its parameters the initial SMILES column and mcsValue. The function can be invoked as described in step 2.

  5. The output of step 4 is a new dataframe scoresDf, its 0-th column contains the scores values. This column, scoresCol, can be reached with by means of byIndex() method of scoresDf.columns object.

  6. Finally, insert the columns mcsCol and scoresCol into the dataframe, next to the position of the initial SMILES column. df.columns.insert() method can help with this, if we cleverly specify the index/position at which the insertion should take place.

Exercise 3: Train Model to Predict Activity​

You will learn: How to train a model inside a package and use it to predict the activity of molecules

Prerequisites: "Molecular fingerprints", "Cheminformatics".

  1. Create a package with the name <yourFirstName>-cheminformatics

  2. Add new function

    // name: TrainAndPredict
    //input: dataframe train
    //input: dataframe test
    export function TrainAndPredict(train, test) {
       // your code here
    }

    Here the training and test dataframes are our datasets for training and prediction, respectively.

  3. Using grok.chem.descriptors create fingerprint of all molecules.

  4. Use grok.ml.trainModel your model (using fingerprint) to predict activity of molecule. You can use dataset example

  5. Using grok.ml.applyModel apply on the test and train datasets. Check the accuracy of the model.

  6. Using grok.shell.addTableView(datasetName) output test dataset