Course Schedule

The full day-by-day plan for the redesigned course. This mirrors course-redesign.md in the repository root, which has the full rationale, licensing notes, and open questions behind this structure. Days already written link to their page; the rest are still to come — the pattern is established on Days 1 and 7 and the remaining days will follow it.

Three phases: bioinformatics foundations build up to profiles (Days 3-6); a convergence zone (Days 7-10) where the classical material directly motivates and is resolved by neural networks; then a deep-learning arc (Days 11-13) that keeps extending the same secondary-structure-prediction thread.

#	Session	Primary reading (open, linked)	Biological case study	Lab
1	Math prerequisites — linear algebra, calculus, probability refresher	d2l.ai, Preliminaries (§2.3 Linear Algebra, §2.5 Calculus, §2.6 Probability & Statistics)	—	None (non-code problem set)
2	Programming prerequisites — Python incl. classes/OOP, then array programming	A Byte of Python (ch. “Object Oriented Programming” for classes); d2l.ai Preliminaries §2.1 Data Manipulation, §2.2 Preprocessing, §2.5 Automatic Differentiation	—	Full
3	Biological Databases	Wikipedia: Biological database, Sequence database; NCBI: GenBank overview; practice via EMBL-EBI Training’s What is ENA? and The UniProt databases	The HBB (beta-globin) locus	Light exercise
4	Sequence Alignment	Wikipedia: Needleman–Wunsch algorithm, Smith–Waterman algorithm, Substitution matrix (BLOSUM/PAM); practice: EMBL-EBI Pairwise Sequence Alignment and Multiple Sequence Alignment	Aligning two homologous globins	Light exercise
5	Advanced Database Search (BLAST)	Wikipedia: BLAST (biotechnology); NCBI: BLAST® Help (public domain)	A real BLAST search on an unknown sequence	Full (covers Days 4+5 together)
6	Profile Methods & HMMs	Wikipedia: Hidden Markov model, Position weight matrix, Sequence profiling tool	Pfam domain search on a protein family — framed explicitly as: a profile/PSSM score is a weighted sum over positions, i.e. already a linear model	Full
7	Linear models, gradient descent & the XOR problem	d2l.ai, Linear Neural Networks for Regression; Wikipedia Perceptron, Perceptrons (book); Bulyk, Johnson & Church (2002), Nucleic Acids Research 30(5):1255-1261	Formalize Day 6’s profile score as a perceptron; show its weights can be learned by gradient descent instead of derived from alignment statistics; then the real-world punchline — non-additive positional dependence in a transcription factor binding site, the same limitation the classic XOR problem illustrates	Full
8	Protein Structure	Wikipedia: Protein structure; RCSB PDB-101	Hemoglobin structure in PyMOL, plus a descriptive preview of Rost & Sander’s early-1990s PHD method	Full (PyMOL)
9	Linear classification (softmax) + practical PyTorch	d2l.ai, Linear Neural Networks for Classification + Builders’ Guide	Subcellular localization prediction; the vehicle for teaching real PyTorch idioms (writing a model as a class, closing the loop with Day 2’s OOP)	Full
10	Multilayer Perceptrons — nonlinearity, backprop, initialization, generalization, dropout	d2l.ai, Multilayer Perceptrons	Payoff of Days 6-9: Rost & Sander’s PHD method end-to-end — profile input → hidden layer → 3-class softmax output → secondary structure	Full
11	CNNs, classic & modern	d2l.ai, Convolutional Neural Networks + Modern Convolutional Neural Networks	Protein contact prediction; SS-prediction’s later profile+CNN era	Full
12	RNNs & LSTM	d2l.ai, Recurrent Neural Networks + Modern Recurrent Neural Networks	SignalP signal-peptide detection; SS-prediction’s (bi)LSTM era	Light exercise
13	Attention & Transformers	d2l.ai, Attention Mechanisms and Transformers	AlphaFold + protein language models — SS-prediction’s endpoint	Light exercise
—	Capstone: Secondary Structure Prediction project	—	Closes the loop opened at Day 8	The project itself