Engineer Trains AI on Drug Database to Find UC Treatments, Uncovers Method's Limits

·3 views

A software engineer with ulcerative colitis built a computational drug repurposing model to test whether machine learning can reliably predict new treatments for the disease. Using the Drug Repurposing Knowledge Graph (DRKG), a database of 97,000 biological entities and 5.9 million relationships compiled by Amazon and several universities, he trained three graph embedding models to score existing drugs as potential UC candidates. The experiment involved removing 108 known UC drug relationships from the dataset, then checking if the models could recover those drugs using indirect biological pathways alone. One model, RotatE, recovered 45% of known UC treatments within the top 100 predictions out of roughly 5,850 compounds, a result that meets published benchmarks. However, the author found that the process also surfaced a drug associated with causing UC, raising deeper questions about whether strong benchmark scores reflect genuine biological insight or simply patterns in how the graph is structured.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Dev Builds Browser-Based Line Break Remover That Handles LF, CRLF, and CR

A developer has built a browser-based line break removal tool called TextFixHub Line Break Remover to address the complexity of handling multiple line-ending formats across platforms. Text copied from sources like Word, spreadsheets, or GitHub can contain LF, CRLF, or CR line endings — sometimes mixed within a single paste. The tool offers three modes: replacing breaks with spaces while preserving paragraphs, removing breaks entirely, or removing breaks while inserting spaces to prevent words from merging. It normalizes all line endings before processing and strips invisible byte-order marks that can silently corrupt text logic. All processing runs locally in the browser with no server uploads, and the tool handles inputs of over 100,000 characters in under 500 milliseconds.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Alchemy GJS Framework Adds WebKit and GStreamer Components for GNOME Apps

Alchemy is a lightweight, reactive UI component library for GNOME desktop applications built with GJS, inspired by the Quasar Framework and Vue.js. The project allows developers to build native GNOME apps using a Vue-like composition API while rendering fully native GTK4 widgets. Two new components have recently been added: QWebView, which integrates WebKit 6.0 with reactive URL and HTML bindings, and QAudioPlayer, a GStreamer-powered audio player with WebRTC support. At its core, Alchemy uses a dependency-tracking reactivity engine modeled after Vue 3, featuring ref, computed, and effect primitives. The framework aims to bridge modern web development workflows with native Linux desktop application development.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

threads=0 Misconfiguration Silently Hangs Waitress-Backed Flask Servers

A common but hard-to-spot misconfiguration in Waitress-backed Flask apps involves setting the threads parameter to 0, leaving the server with no worker threads to process incoming requests. The server starts normally, binds its port, and logs nothing unusual, yet every request hangs indefinitely without any error message. This typically occurs when an environment variable is unset, a config file contains a typo, or a CPU-based thread calculation rounds down to zero on a low-resource container. Because Waitress does not raise an exception for threads=0 at startup, the bug only becomes apparent once requests begin failing silently. The StayPresent library addresses this by validating the threads value before the server starts, immediately throwing a ValueError for any value below 1 to make misconfiguration obvious and easier to debug.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Baking ML model weights into Docker images can be silently undone by one env variable

Developers building serverless GPU applications often embed model weights directly into container images at build time to avoid slow cold starts caused by network downloads. This is done by setting the HF_HOME environment variable to a fixed path inside the image during the Docker build process. However, if any runtime configuration — such as a mounted volume or a template variable — overrides HF_HOME to a different path, the model loader silently falls back to downloading weights from the network, negating the optimization entirely. The failure produces no errors, making it easy to mistakenly blame the hosting platform for slow cold starts. To prevent this, developers are advised to avoid attaching volumes, verify boot logs for unexpected download activity, and explicitly document the cache path as a shared contract between the Dockerfile and the runtime environment.

0 comments Read more at DEV Community