Last week, we were lucky enough to attend the GCC/BOSC conference hosted at Reed. Although I spent the majority of the week volunteering and attending sessions, we also had a deadline for a two-page extended abstract of our CNB-MAC paper to be submitted to the main conference proceedings. We successfully cut down our original paper into two pages and submitted the abstract on Friday.
As of now, we’re almost ready to dig into the experimental portion of this project, but there are still a few things we need to iron out. At the end of last week, Alex noticed a “bug” in how we’ve been calculating the AUC of our multi-layer algorithm. Therefore, one of our priorities this week is to find a more accurate AUC calculation that better represents our method. Once that is completed, we will move on to assembling a list of candidates for experimentally validation. Hopefully by the end of the week we will have chosen 8-10 candidates that are not only biologically interesting but conserved in Drosophila as well.
With our big CNB-MAC paper deadline passed last Friday, my tasks this week have been relatively simple. My first goal this week was to take the bar chart figures in the paper and turn them into box and whisker plots.
For example, I took this figure:
and turned it into this:
The bar chart figure shows the mean AUC of each positive set with varying number of layers, while my figure is a little more descriptive. It shows the mean AUC as well as the interquartile range and outliers.
Next, we got some surprising news! We were notified that after a preliminary pass of the paper submissions, our paper was accepted to the CNB-MAC conference as either a talk or a poster (we’re still waiting to hear which one.) This gave us the opportunity to submit a two-page abstract by the end of next week, a more challenging task. I wrote up a quick rough draft that I will continue working on next week with Alex.
Finally, with our results generated and our paper submitted, it was time to clean up our github repo. This included deleting a lot of old code and useless output files as well as restructuring the code and writing README.md files. It’s looking a lot better now!
This week, my goals have been relatively simple and short. There are a couple of bugs in my code that I need to sort out. I haven’t been able to collect any data about how “good” our method is due to these bugs, but once I do, we will hopefully be able to see that the results we are getting are not random.
Since the last time I wrote a blog post, I have mainly been working on one way of verifying our algorithm; essentially, we need to make sure that our method is actually good at doing what it’s supposed to be doing. Although there are several ways to verify an algorithm, some of which Alex has been working on, I am working on something called “k-fold cross validation.”
This method of verification works by removing (or “hiding”) 1/k-th of your positives, then running your method and seeing where the hidden positives are ranked. You randomly choose the 1/k positives to hide and do this several times. You can also use the distribution of scores you get from each run to see how your method (with all positives) compares – are you getting significant results or are you getting what is expected from randomly choosing positives?
My first attempt at this was just a test run, and there were a few mistakes that I made. First, I chose to sample half instead of choosing a more reasonable number of positives such as 1/4 or 1/5. I also didn’t randomly choose positives to hide – I simply deleted the second half of the positives, and in doing so, might have removed an entire cell motility pathway or two from the positives. This would have produced biased results.
This coming week, my goal is to randomly sample 1/4 of positives multiple times, as well as implement some plotting functions in order to visualize the distribution of scores.
I am taking the biology junior qualifying exam this weekend, so there will be no blog post. The junior qualifying exam is a cumulative exam each student must take their junior year for their major. Each student must pass it in order to be able to move on to writing a thesis their senior year and graduate.
One of our tasks last week was to make a diagram that provides an overview of our project’s methods, pictured below. The blue represents inputs, the yellow represents a function (our semi-supervised method), the green represents outputs, and the purple is the final output.
At the top of the diagram, the blue boxes go through the process of creating the positive list of cell motility genes. The genes were gathered from different signaling pathways in the KEGG database as well as studying primary literature. From these resources, I created two separate positive lists: one purely of cell motility genes and one of cell motility genes implicated in schizophrenia. I combined these together (collapsing the duplicates) in order to create a positive list of 541 cell motility genes. At the bottom of the diagram, the two blue boxes go through the process of creating the schizophrenia positive list. Alex pulled schizophrenia genes from genome-wide association studies (GWAS), then filtered the genes down to the top 300 positives by taking into account their p-values from the literature. We each ran the semi-supervised iterative method on the GIANT brain interactome with the same negative set, differing only in the positive sets we used. From there, the iterative method spit out ranked lists of schizophrenia candidates and cell motility candidates with scores ranging from 0-1 (the green boxes.) Finally, we combined the scores by multiplying them to take into account their probability of being “good” candidates for both cell motility and schizophrenia (the purple box.)
Our runtime-shortening strategies seemed to have worked well! We plotted runtime vs. iteration number and found that each iteration took about 3-4 seconds, so there’s no increase as we saw last time. However, we did find that the scores do not change after a surprisingly small number of iterations, so we will be tweaking a few things to see what changes.
Because it’s hard to tell if a semi-supervised machine learning method is actually good at what it’s supposed to accomplish, we will also be looking into ways to test our method.
Our goals for next time include:
Run on a larger portion of the network. So far we have been using the 0.200 probability threshold network, but we will move down to 0.150 threshold.
Change the number of positives used – see what the effect is of using 1 vs. 300+ positives on the graph.
Plot a distribution of the ranked candidate scores.
Plot the absolute value of the sum of changes made during each iteration.
Look into cross validation to check if our method is doing a good job at what it’s supposed to do. This involves hiding some positives from your positive list and seeing if your method correctly identifies these hidden positives as having a high probability (if not the highest) of being involved in the pathway.
Look into how other papers were able to be convincing about the accuracy of their method.
Next week is spring break so there will be no blog post!
Last week we had the task of running our algorithm for 100,000 iterations. We ran into a couple of problems:
It took a very long time. We actually stopped running it at about 1300 iterations (which took 12 hours) because the estimated time kept increasing. There are a few lines in our code that we are going to change to improve running time – we will convert some variables we’re using to track changes between iterations from lists of nodes to integer counts.
The difference in score we wanted to see before stopping iterations was far too big. We allowed the code to run until the difference in scores between iterations was no more than 0.001 (or it hit 100,000 iterations), and to our surprise, it took around 130 and 180 iterations for cell motility and schizophrenia positives, respectively. Because the scores range from 0 to 1, a change of 10^-4 is bigger than we initially thought. After speeding up the code, we are going to run it until we see changes less than 10^-9.
We are also going to create a histogram of run time versus progress (each iteration) to track our algorithm and see if there is a separate problem that causes the code to slow down as the number of iterations increases.
Our continued goals for this week are to BLAST candidates to see if they are conserved in Drosophila.
We have two goals to accomplish for our meeting next week:
Run our iterative method for 100,000 iterations, or until the scores change by 0.001 or less. See how this affects the ranking compared to 150 iterations. Will it just affect the scores but not the relative rankings?
See if the candidates are conserved in drosophila melanogaster. The second part of this project is to experimentally validate the candidates, which we cannot do if they aren’t conserved in drosophila. As a quick sanity check, we’re going to BLAST CTNNB1 (the well-studied Catenin Beta 1 gene that codes for a protein involved in the formation of adherens junctions) in humans against drosophila which should be ARM.
This previous week we put our iterative algorithm to the test. Alex and I each ran our positive (and negative) lists for 150 iterations using the 0.200 threshold GIANT brain network. As a reminder, the 0.200 threshold network is a trimmed version of the GIANT network where all the edges have probability scores of 0.200 or greater.
I used all 541 of my positives, a combination of two sublists: cell motility genes and cell motility genes implicated in schizophrenia. In the upcoming weeks, I will experiment with running these two lists separately and seeing how the results are affected.
I also used Alex’s list of negatives, which are genes expressed in other organs but not the brain. We will need to closely consider whether it is wise to use this set of negatives alongside cell motility positives.
Even with these two potential issues, our preliminary lists of candidates looked promising – for example, one gene we found is Hes Related Family BHLH Transcription Factor With YRPW Motif 2, also know as HEY2. HEY2 was not only a schizophrenia positive but was also moderately high in the list of cell motility candidates, with a score of 0.56. Another gene we uncovered was 1-Acylglycerol-3-Phosphate O-Acyltransferase 4, or AGPAT4. This was not in either list of positives, but it was given the maximum score of 1.0 in each set of candidates. Both of these genes point to the different ways we can analyze our results and what it means to be a promising candidate – do we care about genes that are positives in the schizophrenia set and are determined by the algorithm to likely be involved in cell motility or do we care about genes that are determined by the algorithm to be likely involved in cell motility and schizophrenia? Do we care about both? What is the difference? Is one more promising than the other? In the upcoming weeks, we will be figuring out these questions in order to refine our list of candidates.
We will also be extending the time that our algorithm runs. 150 iterations may seem like a lot, but cutting the algorithm short may lead to “stunted” results. The iterative method can be thought of as dyes from the positive and negative nodes spreading throughout the network, staining the surrounding nodes in a way that reflects their distances from the positives and negatives. If the iterations are stopped before they converge (i.e. subsequent iterations don’t change the scores, or color of the node), then the color of the nodes might be closer to the color of the positives and negatives than they should be. Our next step will be to let it run for 1500 iterations and see how our results our affected.
Out of my own curiosity, I also decided to see if all of the cell motility positives were in the GIANT network, which turned out to be false. There are about 40 genes (out of 541) that are not in the GIANT network. However, I don’t believe this to be a problem; if there are positive cell motility genes not in the GIANT network, it is probably because not all genes involved in cell motility are expressed in every type of cell and are therefore not relevant in a brain-specific network.
We have a couple of goals for next week: fix a subtle yet major bug in our iterative method program and then run it on our gene lists to get a preliminary list of candidate genes. I will also write a short program to compare the genes in the GIANT network to the list of cell motility genes.