In this repository you will find an analysis of human chromosomes. This analysis primarily focuses on chromosomal morphology. The point of this data science project is to establish whether or not protein composition affects chromosomal alignment and ending structure. Locus locations and amino acid information were used to this end.
In biochemistry, there are many phenomena in which we know little about. Protein formation is one of those. Proteins are the machines that allow our body to function daily. Our body makes proteins everyday, quite frequently. Information about how to make proteins is encoded in DNA. When DNA is organized it is called chromosomes. The only time that DNA organizes into a chromosome is when the cell in which the DNA is located is preparing to divide. All other times, DNA has no discernable structure as far as humans can tell. This chromosomal organization does however provide insight into various types of diseases.
Currently, the only way to track DNA is by radioactively tagging it. This can affect the organism in which the tag is inserted. This study attempts to classify proteins by their location on a chromosome, something that is known, by using the protein's amino acid composition. This amino acid driven classifier could potentially provide an alternative pathway to protein study and isolation for the delivery of drugs.
The type of information being focused on in this project are below in Figure 1.
This tracking system could lead to enhanced drug development capabilities, medicine that is tailored to an individual's own genetic information, a growth in understanding behind chromosomal based diseases, and a variety of other health applications.
I used the data science model for processing data: obtain, scrub, explore, model, and interpret. Ultimately this supervised machine learning model is based on human chromosome 19 and includes only loci that have more than thirty three occurrences, as seen below in Table 1.
| Locus | Occurrences in Chromosome 19 |
|---|---|
| KIR3DL2 | 94 |
| FCAR | 91 |
| KIR2DL4 | 91 |
| NCR1 | 83 |
| KIR2DS4 | 63 |
| KIR2DS2 | 61 |
| CARD8 | 51 |
| RPS9 | 49 |
| OSCAR | 48 |
| ZNF266 | 46 |
| ZNF415 | 46 |
| DMKN | 45 |
| KIR3DL3 | 45 |
| NLRP2 | 40 |
| VSTM1 | 38 |
| KIR2DL1 | 36 |
| CEACAM21 | 35 |
| MBOAT7 | 35 |
| TSEN34 | 34 |
| LILRB1 | 33 |
Table 1. Loci included in machine learning model to study Chromosome 19.
From this collected data, two studies were run. One where loci classification was performed, and a second study, called strand, that identified whether or not the evaluated sequence was a protein.
The models evaluated include: a decision tree classification model, a bagged classification model, and a random forest classification model. A pipeline and a grid search were used to find optimal parameters for the random forest model.
For the loci study, a test accuracy of 99% was achieved using the random forest classifier. For the strand study, the best performing model was the bagged classifier, with a test accuracy of 71%.
Figure 1. Confusion Matrix for best parameters for loci study.
The best parameters for the loci study were:
- max_depth: 9
- min_samples_split: 0.1
- n_estimators: 151.
Figure 2. Confusion Matrix for best parameters for strand study.
The best parameters for the strand study were:
- max_depth: 20
- min_samples_split: 0.1
- n_estimators: 451.
If you have additional questions or would like to discuss any of the contents, feel free to reach out to me either here or through my blog.
The post directly connected to this project can be found here.