Problem Statement:

Based on specific factors, the effectiveness of each medicine in preventing protein development is evaluated. In this experiment, we employed the Inhibitory Concentration (IC 50) value as a marker for analyzing the bioactivity of a specific medicine. The IC 50 value is a quantitative measure of how much a drug is needed to block a given biological process (in this case, multiplication of the EGFR protein) by 50percent. To obtain the canonical notation of a drug’s molecular formulae, we use the ChEMBL dataset, a chemical library of bioactivity molecules. The EGFR constantly ”grows” in cases of lung cancer that tests positive for the EFGR gene due to a gene mutation or malfunction. As a result, cells proliferate out of control the root of which is cancer. Chemotherapy is one of the best ways to treat cancer, but it also has some negative effects, like fatigue, hair loss, and changes in appetite. Therefore, many medications are being investigated to see if they can prevent the EGFR protein from proliferating as an alternative to chemotherapy. Feature selection helps to reduce the dimensionality of the problem, making it easier to work with, and can improve the performance and generalisation ability of machine learning models by reducing the risk of overfitting. In this case there are a lot of features which determine the specifications of a particular molecule which makes it difficult to predict which characteristic might affect a cell from cancer. It plays an important tool for improving the efficiency and effectiveness of machine learning models, especially in problems with highdimensional input data. Training models on all the available characteristics may be computationally expensive and timeconsuming when working with high-dimensional data. We may decrease the computational demands and increase the effectiveness of model training by just choosing the most crucial characteristics. Even when some data is absent, we can still provide good predictions by choosing the most useful characteristics. A lot of data is present in the world regarding the particular protein molecule present in every species of living organisms. It is quite difficult to extract the required targeted molecule which might affect the living being’s conditions of survival.

The methodology for this project was divided into four parts:-

1. Collection of Data.
2. Exploratory Data Analysis.
3. Calculating Molecular Fingerprints.
4. Model construction and feature engineering. A] The data gathering method begins with a search for ’EGFR’ molecules in the ’Chembl’ database. EGFR (epidermal growth factor receptor) is a protein that plays an important role in the regulation of cell growth and division. Mutations in EGFR are associated with a variety of cancers. B] The data is then selected where standard type is IC50. IC50 is a commonly used measure of the potency of a drug compound, and it represents the concentration at which the compound inhibits 50% of the target protein’s activity. Therefore, IC50 values are useful for determining the efficacy of a drug compound in inhibiting a specific target protein. C] Predictors such as molecular weight, molecular LogP value, Number of Hydrogen bone donors and number of hydrogen bone acceptors is calculated from the Lipsinki package. Usefulness of these predictors:
5. 500 Daltons as the molecular weight (MolWt) to ensure that the molecule is small enough to be absorbed and distributed in the body.
6. A partition coefficient (MolLogP) of 5 or less to ensure that the molecule has appropriate lipophilicity and hydrophilicity to cross cell membranes.
7. 5 or fewer hydrogen bond donors (NumHDonors) to avoid excessive hydrogen bonding, which can lead to poor permeability.
8. 10 or fewer hydrogen bond acceptors (NumHAcceptors) to avoid excessive hydrogen bonding, which can lead to poor permeability. D] After calculating the above, pIC50 is calculated. The function pIC50() converts the IC50 values in the input dataframe to pIC50 values (negative log of the molar IC50 concentration), and adds the resulting pIC50 values as a new column to the input dataframe. The standard-value-norm column is used as input to the pIC50() function, which is created in the norm-value() function by replacing any IC50 values greater than 100000000 with 100000000. E] The next phase involves the conversion of the canonical formula into its corresponding molecular fingerprint with the help of the other descriptors. F] The program standardizes the data by removing any extraneous salts or chemical components from the molecule. G] The algorithms and methods we are looking forward to implement use are Logistic Regression, Ada-Boost Regressor, Bagging Regressor and Random Forest Regressor. Our project aims to compare the methods and best models out there to to predict the efficiency of a therapy in preventing the development of cancer cells caused by a deficiency in EGFR protein.

EDA:

Experiments and Results The following processing and predictors resulted in the following: 1. There are more number of active molecules than inactive molecules. This is calculated with the help of standard value. 2. The ratio between molecular weight and its LogP for active and inactive molecules is somewhat same. 3. The pIC50 value should be less than 6 for the molecules to be inactive. 4. The distribution of molecular weight and its LogP for both inactive and active molecules is same. But they have different medians. Inactive molecules have higher median which is a threat. 5. The number of hydrogen bond donors are same for both active and inactive class whereas there are more hydrogen bond donors are higher for active class.