Personalized Cancer_Diagnosis

1. Task: Classify the given genetic variations/mutations based on evidence from text-based clinical literature

1.1 In detail explaination:

1. Lets take a example where a person seems to have a symptoms of cancer and he visits a hospital where they basicaly remove the Cancer Tumor from person & then sequence those cancer tumor (genetic sequencing)
2. So after doing genetic sequencing of cancer tumor we get gene, gene mutation/variations where a small variation in a gene can cause cancer by destroying entire genetic core system
3. Note: All types of Mutation does not cause cancer only few of them cause
4. So this is what we are going to do with help of features (gene , gene mutation ) we will predict 9 diffrent class labels and determine that out of which actually cause Cancer.

1.2. Workflow :

1. Lets take a domain expert who understand about gene mutation, selects list of [genetic varaiations] he want to analyize. Genetic variations divide into 2 parts (1: gene {on which variation is happening} 2: varaiation {exact varations}) Both can be considered as [Random Categorical Variables]
2. Domain Expert will now search/ collect all research papers/ evidence/ text that has been done on this gene variations
3. Spend some time on analyzing Text [i.e. Evidence / Reseach work] amd after that Determine that patient belongs to which of the class out of (1,2,3,4....,9)

Information gathered from: https://www.kaggle.com/c/msk-redefining-cancer-treatment/discussion/35336#198462

1.3 Data overview:

Source: https://www.kaggle.com/c/msk-redefining-cancer-treatment/data

1. Training variants (ID,Gene,Variations,Classlabel)
2. Training_Texr (ID, TEXT)

1.4 Type of Machine learning problem

There are nine different classes a genetic mutation can be classified into => Multi class classification problem

1.5 Performance Metric

1. Multi-class Log loss
2. Confusion matrix

2. Exploratory Data Analysis

2.1 Distribution of Train,Test,Cv data (64:20,16)

Distribution of yi in Train

Distribution of yi in Test

Distribution of yi in CV

Observation: Here we have done Stratified Train Test CV split so that we get equal amount of distribution in all 3 plots as we can see and by looking at this Histogram we can come into concluison that Class Label {7,4,1,2} consist of more than 60% - 70% of data and so this is an Imbalanced Dataset

2.1 Univariate Analysis

2.1.2 Univariate Analysis on Gene feature

Q.How many categories are there and How they are distributed?

unique_genes = train_df["Gene"].value_counts()

(240,)

Plot CDF for this distrbutions of gene feature

s = sum(unique_genes.values)
h = unique_genes.values/s
c = np.cumsum(h)
plt.plot(c,label='Cumulative distribution of Genes')
plt.grid()
plt.legend()
plt.show()

Observation: Out of 225 unique genes only top 50 genes consist more than 70% of data and rest only 30%

2.1.3 Univariate Analysis on Variation feature

Q.How many categories are there and How they are distributed?

unique_genes = train_df["Variation"].value_counts()

(1941,)

Plot CDF for this distrbutions of gene feature

s = sum(unique_variation.values)
h = unique_variation.values/s
c = np.cumsum(h)
plt.plot(c,label='Cumulative distribution of Genes')
plt.grid()
plt.legend()
plt.show()

3. Machine Learning models used

1. Multinomial Naive Bayes

2. K- Nearest Neigbour

3. Logistic Regression with Class Balancing

4. Logistic Regression without Class Balancing

5. Linear SVM