Feature Preprocessing

The range of values for a given feature can impact an algorithm's performance

Remember using the value of the year directly on assignment 1?

Linear scaling

Scale the values into the range $[0,1]$

$x \leftarrow \frac{ x - x_{min} }{ x_{max} - x_{min} }$

Scale based on training set only

Z-normalization

Scale the distribution to have mean=0 and std=1

$x \leftarrow \frac{ x - \mu_X }{ \sigma_X }$

Scale based on training set only

Feature Discretization

Some algorithms only work on discrete features

We may need to discretize real-valued features

Feature Discretization

Calculate the histogram

This divides the values into bins

• Equal bin sizes
• Equal # of instances per bin

Feature Discretization

Alternatively, use a heuristic/ad-hoc method to discretize in a useful way

E.G. Build a decision tree, let the DT algorithm discretize, and use the split values of the optimized tree

From Discrete to Numerical

Some features are unordered (i.e. Browsers = [ Firefox, Chrome, Safari ])

Most common approach is to use unit vectors:

Final Project

Proposal due: 03/07

Write a 300-500 word abstract describing your proposed project. This should include 2-3 references of papers you expect to include in your final paper.

Final Project

Project due: 04/25

Turn in a 8-12 page paper. A rough outline is:

• Background on problem
• Related work
• Your method
• Results
• Conclusion and future work
• References

Final Project: Being prepared

• Take excessive amounts of notes (Powerpoints are good for note taking)
• Keep track of sources
• Track down datasets/verify they can be accessed
• Jupyter can be used for writing text, doing ML, and making figures/graphs

Final Project: Resources

• Google Scholar, Sci-hub for references
• Word, Libreoffice, Latex/Overleaf, Markdown+Pandoc for writing
• Store code and even the paper on Github, Bitbucket, Gitlab
• Notebook style coding/writing (i.e. Jupyter) will help form the good habit of excessive note taking

Background on problem

B-cell receptors bind specific molecular structures to recruit immune system activity.

Background on problem

Background on problem

The epitope prediction problem:

Given that a protein is a sequence of amino acids that fold into a 3D shape, only a subset of these amino acids are bound by B-cell receptors. Which ones?

Exciting Title

Complex Features in Prediction of Discontinuous B-cell Epitopes

This is a project with James Chin.

Related work

There are existing methods for B-cell epitope prediction. They vary in terms of specific feature sets, machine learning algorithms, and training/test sets.

• Discotope - Log-odds ratios for features (Haste Anderson et al, 2006)
• Epitopia - Naive Bayes classifier (Rubinstein et al, 2009)
• PEASE - Random forest (Sela-Culang et al, 2014)

Method

Much of the advancement in B-cell epitope prediction has been driven by improvement in features.

We propose a method for synthesizing complex features from a set of basis features that allows provides high-degrees of nonlinearity not possible with simple ML representations.

Results

References