Nucleobase Counting
Contents
Motivation
- get our feet wet with parallel programming
- gain experience with the async and finish constructs
- take different approaches to splitting up the work
We will solve the problem sequentially, then split the work up into 2 tasks, then coarsen the work n-ways, and finally split up the work in a divide-and-conquer recursive style.
Background
Bioinformatics
For this assignment, you will be writing sequential and parallel code to count nucleobases in a human X chromosome.
DNA is made up of four nucleobases: cytosine, guanine, adenine, and thymine. A strand of DNA can thus be represented as a string of letters representing these nucleobases, for example: “ACCGCATAAAGTCC.” However, DNA sequencing is typically not 100% accurate, so some of the nucleobases are not read with high certainty. These bases can be represented as an “N.” A sequence then might look something like “NCCGCATNAAGTCC.” Your goal is to write code that counts the number of occurrences a particular nucleobase or uncertain reads.
We will be using actual data pulled from the US National Library of Medicine, a database maintained by the National Institute of Health. We have already provided you the code that you need to access the chromosome from the database and check your work. You must implement a sequential solution and three parallel solutions to count the given bases in these sequences.
For some more optional background on DNA and nucleotide bases, please check out
Parallel Programming
Parallel Constructs
- finish
- async
- Feel free to use the course reference page if you need a reminder on how to use asyncs and finishes
Related Videos
The Core Questions
- What are the tasks?
- What is the data?
- Is the data mutable?
- If so, how is it shared?
Mistakes to Avoid
Warning: Be sure to remove each NotYetImplementedException as you implement your solutions. |
Warning: Do NOT copy the data. We are simply reading the chromosome data, so there is no reason to copy it. |
Warning: Do NOT share data beyond what is necessary. Do NOT use static class fields when you could use local variables instead. |
Warning: Do NOT have the same for loop code duplicated throughout your solution. Invoke countRangeSequential where appropriate. |
Warning: Do NOT not invoke countSequential to count the entire chromosome by mistake when you want to invoke countRangeSequential to count the range from [min,max). |
Getting Started
Code to Implement
Midpoint
class: | MidpointUtils.java | |
methods: | caclulateMidpoint | |
package: | midpoint.lab | |
source folder: | student/src/main/java |
method: public static int calculateMidpoint(int a, int b)
(sequential implementation only)
In this method, you will need to calculate the midpoint between two numbers. This is as simple as finding the average between the two numbers. There is no need to worry about rounding correctly, just drop everything after the decimal point if the midpoint is not automatically an int.
There are at least two correct ways to implement the midpoint:
- Option A:
- Option B:
It is hard to find 1D references for midpoint. Wolfram MathWorld has a breakdown of Midpoint in 2D and 3D. Pick any dimension you like.
Threshold
class: | GreaterThanThresholdPredicate.java | |
methods: | constructor test |
|
package: | threshold.lab | |
source folder: | student/src/main/java |
method: public GreaterThanThresholdPredicate(int threshold)
(sequential implementation only)
method: public boolean test(int value)
(sequential implementation only)
Class GreaterThanThresholdPredicate implements Java IntPredicate interface. The test method should return a boolean value that indicates whether the input argument is greater than the threshold provided in the class constructor.
Count Range
method: public static int countRangeSequential(byte[] chromosome, Nucleobase targetNucleobase, int min, int maxExclusive)
(sequential implementation only)
This utility method should count the number of times a particular nucleobase occurs in the array between [min, maxExclusive).
Sequential Solution
method: public static int countSequential(byte[] chromosome, Nucleobase targetNucleobase)
(sequential implementation only)
This solution should be achieved with a simple call to one of the utility methods you wrote.
Parallel Solutions
You need to implement three different parallel solutions to this problem. The first will involve splitting the array into two equal halves, then going through each half of the array in parallel. The second will involve splitting the array into n different pieces, then going through each of those pieces of the array in parallel. The third and final implementation will recursively divide the array until a threshold is reached.
Lower/Upper Split
method: public static int countParallelLowerUpperSplit(byte[] chromosome, Nucleobase targetNucleobase)
(parallel implementation required)
In order to start with then two equal halves solution, please modify the countParallelUpperLowerSplit
method.
Coarsening N-Way Split
method: public static int countParallelNWaySplit(byte[] chromosome, Nucleobase targetNucleobase, int numTasks)
(parallel implementation required)
After you’re finished with that, please modify the countParallelNWaySplit
method for splitting the work up into n tasks solution.
If the chromosome was 1,000,000 long, and you were asked to split it up in 10 tasks, each task should obviously process 100,000. Things get slightly tricky if the chromosome were 1,000,005 long.
Divide and Conquer
In a pattern that we will use repeatedly in the course, we have a public method that is asked to count an entire chromosome. This method just calls the kernel method to do all of the work:
public static int countParallelDivideAndConquer(byte[] chromosome, Nucleobase targetNucleobase, IntPredicate isParallelRangeLengthPredicate) throws InterruptedException, ExecutionException { return countParallelDivideAndConquerKernel(chromosome, targetNucleobase, isParallelRangeLengthPredicate, 0, chromosome.length); }
method: static int countParallelDivideAndConquerKernel(byte[] chromosome, Nucleobase targetNucleobase, IntPredicate isParallelRangeLengthPredicate,
int min, int maxExclusive)
(parallel implementation required)
The kernel method will recursively divide the work up in parallel until the range drops below the threshold specified by isParallelRangeLengthPredicate. At that point it will simply use the sequential count range utility method.
Testing Your Solution
How to effectively use JUnit:
Correctness
class: | CountTestSuite.java | |
package: | count.lab | |
source folder: | testing/src/test/java |
Launch CountTestSuite.java as a JUnit Test to run all of the tests. CountTestSuite.java is located in src/test/java count.lab package. You can initiate this via right clicking on CountTestSuite.java and selecting "Run As..." -> "JUnit Test". [screenshot]
Performance
class: | NucleobaseCountTiming.java | |
package: | count.lab | |
source folder: | src/main/java |
After your JUnit tests are passing and you are confident that your solutions are correct, move on to checking out the timing performance of your different solutions with different task counts.
Pledge, Acknowledgments, Citations
file: | lab-count-pledge-acknowledgments-citations.txt |
More info about the Honor Pledge
Rubric
As always, please make sure to cite your work appropriately.
Total points: 100
- Correct countRangeSequential (10)
- Correct countSequential (10)
- Correct calculateMidpoint (5)
- Correct countParallelLowerUpperSplit (10)
- Parallel countParallelLowerUpperSplit (10)
- Correct countParallelNWaySplit (15)
- Parallel countParallelNWaySplit (10)
- Correct GreaterThanThresholdPredicate (5)
- Correct countParallelDivideAndConquerKernel (15)
- Parallel countParallelDivideAndConquerKernel (10)
Penalties may be assessed for clarity and efficiency.