MIT – Computational Systems Biology: Deep Learning in the Life Sciences

MIT – Computational Systems Biology: Deep Learning in the Life Sciences

Course description

This courses introduces foundations and state-of-the-art machine learning challenges in genomics and the life sciences more broadly. We introduce both deep learning and classical machine learning approaches to key problems, comparing and contrasting their power and limitations. We seek to enable students to evaluate a wide variety of solutions to key problems we face in this rapidly developing field, and to execute on new enabling solutions that can have large impact. As part of the subject students will implement solutions to challenging problems, first in problem sets that span a carefully chosen set of tasks, and then in an independent project. Students will program using Python 3 and TensorFlow 2 in Jupyter Notebooks, a nod to the importance of carefully documenting your work so it can be precisely reproduced by others.

Syllabus and schedule

	When	Where	Description	Course materials	Reference
Lecture 1	Feb 16 1pm	Course Intro + Overview Foundations	Read Goodfellow Chapter 1Lecture slidesLecture video	DL in BioinformaticsDL for computational biologyThe Roots of BioinformaticsML in Genomic MedicineAwesome DeepBioVisual Information Theory
Lecture 2	Feb 18 1pm	ML Foundations	Read Goodfellow Chapter 6Feed Forward BackpropLecture slidesLecture video	DamienNick LocascioTF site tutorials
Recitation 1	Feb 19 3pm	ML Review	Recitation slides
Lecture 3	Feb 23 1pm	Convolutional Neural Networks	Read Goodfellow Chapter 9Lecture slidesLecture video
Lecture 4	Feb 25 1pm	Recurrent Neural Networks, Graph Neural Networks	Read Goodfellow Chapter 10Lecture slidesLecture video
Recitation 2	Feb 26 3pm	Neural Networks Review	Recitation notes
Lecture 5	Mar 2 1pm	Interpretability, Dimensionality Reduction	Lecture slidesLecture video	Binder et al. (Relevance Propagation)Dumoulin and Visin (Convolution Arithmetic)Finnegan and Song (Maximum entropy methods)Lundberg and Lee (SHAP)Ribeiro (LIME)Selvaraju et al. (Grad-CAM)Shrikumar et al. (Learning Important Features)Shrikumar et al. (DeepLIFT)Simonyan et al. (Saliency Maps)Springenberg et al. (CNN)Sundararajan et al. (Axiomatic Attribution)Yosinski et al. (Deep Visualization)Zeiler et al. (Deconvolutional Networks)Zeiler and Fergus (Understanding Convolutional Networks)Zhou et al. (Discriminative Localization)
Lecture 6	Mar 4 1pm	Generative Models, GANs, VAE	Lecture slidesLecture video
Recitation 3	Mar 5 3pm	Interpreting ML Models	Recitation slides
No class	Mar 9		Monday Class Schedule
Deadline	Mar 10 11:59pm		PS1 due
Lecture 7	Mar 11 1pm	DNA Accessibility, Promoters and Enhancers	Lecture slidesLecture video
Recitation 4	Mar 12 3pm	Chromatin and gene regulation	Recitation slides
Lecture 8	Mar 16 1pm	Transcription Factors, DNA methylation	Lecture slidesLecture video
Lecture 9	Mar 18 1pm	Gene Expression, Splicing	Lecture slidesLecture video
Recitation 5	Mar 19 3pm	RNA-seq, Splicing	Recitation slides
No class	Mar 23		Class Holiday
Lecture 10	Mar 25 1pm	Single cell RNA-sequencing	Lecture slidesLecture video
Recitation 6	Mar 26 3pm	scRNA-seq, dimensionality reduction	Recitation slides
Lecture 11	Mar 30 1pm	Dimensionality Reduction, Genetics, and Variation	Lecture slides ALecture slides BLecture video
Lecture 12	Apr 1 1pm	GWAS and Rare variants	Lecture slidesLecture video
Deadline	Apr 1 11:59pm		PS2 due
Recitation 7	Apr 2 3pm	Genetics	Recitation slides ARecitation slides B
Lecture 13	Apr 6 1pm	eQTLs	Lecture slidesLecture video
Lecture 14	Apr 8 1pm	Electronic health records and patient data	Lecture slidesLecture video
Recitation 8	Apr 9 3pm	ML for health data	Recitation slides
Lecture 15	Apr 13 1pm	Graph analysis	Lecture slides Part ALecture slides Part BLecture video
Lecture 16	Apr 15 1pm	Drug discovery	Lecture slidesLecture video
Recitation 9	Apr 16 3pm	Protein structure prediction	Recitation slides
No class	Apr 20		Class Holiday
Lecture 17	Apr 22 1pm	Protein folding	Lecture slides Part ALecture slides Part BLecture slides Part CLecture video
Deadline	Apr 23 11:59pm		PS3 due
Recitation 10	Apr 23 3pm	Exam prep session		Recitation slides
Exam	Apr 27 11:59pm		In-class exam
Lecture 19	Apr 29 1pm	No lecture
Deadline	Apr 29 11:59pm		PS4 due
Recitation 11	Apr 30 3pm	Structural biology and protein folding	Recitation slides
Lecture 20	May 4 1pm	Imaging applications in healthcare	Lecture slides Part ALecture slides Part BLecture video
Lecture 21	May 6 1pm	Video processing, structure determination	Lecture slidesLecture video
No class	May 7		Class Holiday
Lecture 22	May 11 1pm	Imaging and Cancer	Lecture slides Part ALecture slides Part BLecture video
Lecture 23	May 13 1pm	EHRs and data mining	Lecture slidesLecture video
Recitation 12	May 14 3pm	How to present	Recitation video
Deadline	May 17 11:59pm		Final project reports due
Lecture 24	May 18 1pm	Neuroscience	Lecture slides Part ALecture slides Part BLecture video
Deadline	May 19 11:59pm		Final presentations due
Deadline	May 20		In-class final presentations

Tutorials for TensorFlow, NumPy, Google Cloud, and Jupyter notebooks

We collected a series of pointers to tutorials on NumPy, TensorFlow, Google Cloud and Conda here. We also provide a Quickstart tutorial to set up essential environment and tools for you to work on problem set 0 and problem set 1.

Prerequisites

You should be comfortable with calculus, linear algebra, (Python) programming, probability, and introductory molecular biology. This will be a fast paced course, and it is targeted towards students that are both mathematically and computational capable. There are many other subjects at MIT that teach overviews of computational biology that are less demanding, we would be happy to recommend other options if you find this subject is not what you desire.

Class meeting times

• Lecture: TR1-2.30
• Recitation: F3-4
• Mentoring Session: F4-5

Contact

You should feel free to contact the lecturer and the TAs about any questions through 6.874staff@mit.edu. The best way to get detailed questions answered is to attend TA office hours and recitation or post them on Piazza.

Office hours

Manolis Kellis (manoli@mit.edu): M 5-6pmZheng Dai, Dylan Cable: Tues 4-5pmJackie Valeri, Tessa Gustafson: Wed 7-8pm

Grading

Grading will be based upon five programming-intensive problem sets (30%), a quiz (25%), a project (35%), and participation plus one day of lecture scribing (10%). Attendance in lecture is important as the class moves quickly and you will need to be present. For students enrolled in one of the graduate versions of this class (6.874, 20.490, and HST.506) there will be an extra section on some problem sets. You can use three late days for problem set deadlines (or email the course staff).

Lecture Scribing

If you are enrolled in this course for credit, you are requiured to scribe for one lecture.

The requirements for lecture scribing are as follows:

1. On the day of lecture you may take notes however you like. Lectures will be recorded, so asynchronous participation is fine.
2. During the week after lecture, we ask that you work with everyone assigned to scribe your lecture to compile a finalized set of notes that summarize the key points of the lecture, explain important equations, images and plots, illustrate or describe relevant things that were written on the board, and describe any important questions & answers between student and professor that were exchanged.
   The end goal is for you to generate a compact resource which you and your classmates can use to glean the important material from your lecture. The finalized notes should generally adhere to and extend from the structure outlined by the headings at the beginning of the notes template.
3. The notes template and finished scribed notes may be found here.
4. Let the course staff know you are finished compiling the notes by sending an email to 6.874staff@mit.edu. The deadline for completing the notes will be end-of-day one week after your lecture (e.g. notes from a lecture on 2/18 will be due on 2/25 @ 11:59 PM).

Project

This subject has a substantial project component. We strongly recommend working on projects in team of 2-3 students, but if there’s a strong justification, we can consider exceptions. You are free to choose any problem in the life sciences related to the lectures of the course, and develop a deep learning solution using the subject’s methodologies or cloud resources. We will have extensive mentoring resources for the students to help provide guidance, access to datasets, and biological insights. We will hold mentoring sessions during which you will have a chance to refine your ideas in consultation with the teaching staff and research mentors for each research area.

Textbook

We will be using the book “Deep Learning” by Goodfellow, Bengio, and Courville. You can find the book online here and here. You can purchase a hard copy at MIT Press or on Amazon.

Another useful book is the Matrix Cookbook, an extensive collection of facts about matrices.MIT – 6.802 / 6.874 / 20.390 / 20.490 / HST.506 Computational Systems Biology: Deep Learning in the Life Sciences – Spring 2019