Concept
Patterns, not programs
Instead of hard-coding rules, ML fits functions to examples and generalizes to new data.
Elton Ugbogu
Concise Ideas. Clear science.
A space for exploring machine learning, bioinformatics, and cancer biology — with room for playful curiosity
Quick look
What's inside
- Plain-English ML concepts: how models learn patterns (and avoid overfitting).
- How sequencing data becomes features you can actually learn from.
- Understanding the intricate world of cellular dysfunction and growth.
Machine Learning — the tiny tour
Practice
Train · Validate · Test
Split wisely; tune on validation; keep a clean test set. Reliable evaluation beats lucky results.
Intuition
Bias–Variance balance
Too simple misses patterns; too complex memorizes noise. The sweet spot generalizes best.
Bioinformatics — data into discovery
From raw reads to tidy tables: QC → alignment → quantification → normalization → interpretation.
- Features: genes, transcripts, variants, pathways.
- Normalization: methods like TPM, VST, and TMM make samples comparable.
- Reproducibility: workflows keep results reliable and shareable.
Friendly glossary
- Feature matrix: rows = samples; columns = features.
- Label: the thing we predict (e.g., tumor type).
- Cross-validation: repeated splits to estimate real-world performance.
Cancer Biology — unraveling cellular complexity
Idea
Hallmarks of Cancer
Understanding shared characteristics like uncontrolled proliferation and evasion of growth suppressors.
Research
Genomic & Epigenomic Shifts
Investigating mutations, copy number variations, and epigenetic modifications driving cancer development.
Therapy
Targeting Pathways
Developing treatments that block specific molecular pathways crucial for cancer cell survival and growth.
Current Advances in Machine Learning
From foundation models to self-supervised learning, ML continues to push boundaries across various domains.
- Generative AI: Large Language Models (LLMs) and diffusion models creating new content.
- Reinforcement Learning: Agent training in complex environments, like robotics and game playing.
- Explainable AI (XAI): Making black-box models more transparent and interpretable.
Frontier Applications
- Drug Discovery: Accelerating identification of new compounds and targets.
- Personalized Medicine: Tailoring treatments based on individual patient data.
Quantum ML — a curious frontier
Quantum ideas (like superposition and entanglement) inspire models that may help with small, high-dimensional data. Hardware is early, but the playground is exciting.
Where it's heading
- Kernel methods and hybrid models are popular starting points.
- Early studies explore materials, chips, and complex physical systems.
- As devices improve, expect more practical bio/health demos.
Selected writing
Featured essay
Tumour Ecosystem Dynamics
A long-form overview of malignant cells, stromal and immune compartments, spatial transcriptomics, deconvolution, and the theoretical limits of tumour microenvironment inference.
Featured essay
Maximum entropy models in biology and machine learning
An integrated reference on Jaynes's principle, exponential-family models, Ising and Potts systems, protein coevolution, gene-network inference, and statistical mechanics.
Archive
Writing archive
Browse all published long-form essays in one place, including machine learning, bioinformatics, cancer biology, and quantum-inspired work.
Visualizing the Future
About
I'm studying Bioinformatics at Universität Potsdam and I love machine learning.
Find me
Contact
Say hello: contact@eltonugbogu.com