Pavel-Kravchenko

scRNA-seq dimensionality reduction and clustering using PCA, k-NN graph, UMAP, and Leiden. Includes a guide for parameter selection, implementation patterns, and common pitfalls.

In silico restriction digestion, compatible end detection, primer design (Tm models), and gel simulation.

Viral genome assembly, intra-host variant calling, phylodynamics, and real-time surveillance.

Vision-language model inference patterns for scientific documents.

Genomic Foundation Models: Nucleotide Transformers, HyenaDNA, and Evo with NumPy.

A self-balancing BST (Adelson-Velsky & Landis, 1962) guaranteeing O(log n) operations via rotation-based rebalancing.

Dijkstra's Algorithm: Shortest Paths in Weighted Graphs

Suffix trees are compressed tries of all suffixes, enabling O(m) pattern search and O(n) construction via Ukkonen's algorithm.

ESM2 Embeddings and ESMFold with NumPy

Module T5-01B: LLM Training Systems (Tracking, Epochs, and Ablations) with Pandas

A singly linked list with a full implementation, featuring head/tail pointers, operations for insert, delete, search, and reverse, along with a complexity table.

ATAC-seq quality control and accessibility analysis.

Molecular Modeling with NumPy

Long Non-Coding RNA: Discovery and Classification

AlphaFold2/3 architecture, confidence metrics (pLDDT/PAE), structure fetching, and protein design ranking workflows.

Score matching, noise schedules, DDIM sampling, and DDRM inverse problems for diffusion generative models.

Enformer architecture for regulatory prediction from DNA, in-silico mutagenesis (ISM), and variant prioritization.

Borzoi vs Epiformer model selection for RNA-seq coverage vs chromatin accessibility prediction.

Geneformer and scGPT for Single-Cell Modeling.

LLM fine-tuning with HuggingFace Transformers, covering LoRA math, QLoRA 4-bit NF4 quantization, chat templates, and the SFTTrainer workflow.

Splicing Models: SpliceAI and AlphaGenome with NumPy

From DNA Variants to Protein Structure: AlphaFold2, AlphaFold3, RoseTTAFold with NumPy

Module T5-02: Vision RAG with NumPy

Zero-Shot Mutation Effect Prediction with NumPy

Multi-pattern string matching in O(n + m + z) via a trie augmented with KMP-style failure links.

Foundational algorithms — Euclidean GCD, Newton's method for cube root — with complexity analysis and clean implementations.

BST operations and complexity, with a clean implementation using parent pointers to support all standard operations.

Comparison-based sorting algorithms — bubble, merge, shell, quicksort — with complexity tables and implementation patterns.

Graph representations — adjacency matrix, adjacency list, edge list — with complexity tables, implementation patterns, and biological network use cases.

Hash tables (chaining vs open addressing) and Bloom filters: complexity, trade-offs, and implementation patterns.

Top-down DP with memoization: caches recursive subproblems for an exponential-to-linear speedup, using the @lru_cache shortcut.

KMP string matching: O(n+m) pattern search using a failure function prefix table.

Knapsack DP variants, including 0/1, unbounded, and subset sum, with traceback and space optimization.

Covers linear and binary search implementations, variants (first/last occurrence), and their respective use cases.

Counting sort, radix sort, and bucket sort are O(n) non-comparison sorts that bypass the Omega(n log n) lower bound.

Naive O(nm) brute-force string matching — sliding window baseline, when acceptable, motivation for KMP/Rabin-Karp.

Rabin-Karp hash-based string matching uses a rolling hash for O(n+m) average time complexity and excels at multi-pattern search.

Red-black tree: a self-balancing BST with O(log n) operations, maintaining 5 invariants, and fixing insertions with rotations and recoloring.

Needleman-Wunsch global and Smith-Waterman local sequence alignment, including DP table fill, traceback, and scoring matrices.

Stack (LIFO) and Queue (FIFO) — array and linked-list implementations, O(1) operations, common applications.

Suffix arrays are a space-efficient sorted suffix index, offering O(n log n) construction and O(m log n) pattern search.

Bottom-up Dynamic Programming with tabulation, covering edit distance, LCS, and space optimization using rolling arrays.

Topological sort using DFS and Kahn's BFS algorithms for DAG ordering, cycle detection, and critical path analysis.

Prefix tree for O(m) string insert/search and O(p+k) prefix queries; ideal for autocomplete and dictionary membership.

AlphaFold/ESMFold structure prediction and confidence interpretation.

Advanced NGS: genome assembly algorithms (OLC, de Bruijn graph), k-mer theory, assembler selection table, and SPAdes/Flye/hifiasm CLI usage.

Metagenomic assembly with MEGAHIT, contig binning with MetaBAT2, and MAG quality assessment with CheckM. Includes binning signals, multi-sample strategy, and MIMAG quality tiers.

Bayesian statistics with PyMC: prior specification, MCMC sampling, posterior analysis, and hierarchical models for biological data. Use when applying Bayesian inference to experiments.