A team of researchers recently fed 9,555 natural compounds into an AI screening pipeline and out popped a molecule from the ginkgo tree that appears to fix broken cellular garbage trucks in motor neurons. Standard Tuesday, really.

According to a study published in Autophagy by Ang Li and colleagues at the University of Macau, Guangzhou Medical University, and Imperial College London, the compound isoginkgetin - a biflavonoid hiding in Ginkgo biloba leaves since roughly the Jurassic period - activates a protein called TFEB that controls lysosomal biogenesis. Translation: it flips the switch that tells cells to build more lysosomes, the tiny recycling centers responsible for breaking down cellular waste. And in amyotrophic lateral sclerosis (ALS), those recycling centers are catastrophically broken (Li et al., 2026).

The Cellular Garbage Crisis Nobody Talks About

ALS destroys motor neurons - the nerve cells that connect your brain to your muscles. Most people know that part. What fewer people realize is that one of the earliest things to go wrong isn't the neurons themselves but their internal waste management system.

Your cells constantly produce junk: misfolded proteins, damaged organelles, molecular debris. Lysosomes are the industrial shredders that process all of it. TFEB is the foreman who decides when to build more shredders. In ALS, TFEB gets stuck on the lysosomal membrane, phosphorylated into silence by enzymes like mTORC1 and GSK-3β. The result? Toxic protein aggregates - particularly TDP-43 and SOD1 - pile up like uncollected trash after a city workers' strike (Xia et al., 2022; Root et al., 2021).

The numbers tell a grim story. Effective pharmacological strategies to restore lysosomal homeostasis? According to the researchers: "remain limited." That's science-speak for "we basically have nothing."

Enter the Algorithm

Here's where the investigation gets interesting. Rather than testing thousands of compounds one petri dish at a time - the pharmaceutical equivalent of finding a needle in a haystack by hand - Li's team deployed machine learning-based virtual screening to computationally sift through a library of 9,555 natural compounds. The AI was hunting for molecules that could activate TFEB by inhibiting GSK-3β, the enzyme that keeps TFEB phosphorylated and trapped outside the nucleus.

The system flagged isoginkgetin. When pressed against the molecular evidence, the compound turned out to be an ATP-competitive inhibitor that binds directly to the Lys85 residue in GSK-3β's ATP-binding pocket. With GSK-3β neutralized, TFEB shakes off its phosphate chains, migrates to the nucleus, and starts ordering the construction of fresh lysosomes like a factory manager whose budget just got approved.

This is the kind of result that makes drug discovery researchers sit up straight. AI-driven screening is reshaping how we find therapeutic candidates, and this study is a clean example of the approach working exactly as advertised - with the added bonus that the hit compound comes from a tree that's been around for 270 million years.

The Evidence Trail

But a computational prediction is just a hypothesis. The real question: does isoginkgetin actually protect motor neurons?

The same research group has been building a comprehensive case across multiple studies. In a companion paper published in EMBO Molecular Medicine, isoginkgetin demonstrated neuroprotective effects across an unusually broad range of ALS models: C. elegans worms, SOD1-G93A transgenic mice, and - critically - motor neurons derived from actual ALS patient stem cells carrying three different genetic mutations (C9orf72, SOD1, and TDP-43). That last detail matters enormously because ALS isn't one disease but a family of related conditions with different genetic triggers (Li et al., 2025).

The mechanism appears to work through at least two complementary pathways. The TFEB-lysosomal biogenesis pathway (the focus of the Autophagy paper) ramps up the cell's garbage processing capacity. A parallel PINK1-Parkin mitophagy pathway specifically targets damaged mitochondria for disposal. It's a two-pronged cleanup crew.

Why Your Skepticism Is Healthy (But So Is the Data)

A compound from a "living fossil" tree, identified by AI, that rescues motor neurons through multiple mechanisms? If your internal alarm for "too good to be true" just fired, good - that instinct serves you well in science journalism.

But dig into the supplementary materials and a few things stand out. The protective effects disappeared when researchers knocked out the target pathways (PINK1 or Parkin), confirming the mechanism isn't some vague, general antioxidant hand-waving. The cross-species and cross-mutation validation is unusually thorough for a preclinical study. And the AI screening approach is transparent and reproducible.

The caveats are real, though. This is preclinical work. The gap between "protects motor neurons in a dish" and "helps people with ALS walk" is roughly the size of the Grand Canyon, and the pharmaceutical graveyard is full of promising ALS drug candidates that looked great until they didn't. Only about 3-5% of ALS drug candidates that enter clinical trials actually make it to approval.

The Bigger Picture

What makes this study genuinely noteworthy isn't just the molecule - it's the methodology. AI-based virtual screening is increasingly being used to identify drug candidates for neurodegenerative diseases, from repurposing existing medications to discovering entirely new compounds. A recent analysis of U.S. veteran health records used machine learning to identify 27 existing medications associated with longer ALS survival - a completely different approach to the same devastating problem.

If you've ever tried to visually map the tangled web of cellular pathways involved in neurodegeneration, tools like mapb2.io can help you sketch out these complex signaling cascades - because honestly, the GSK-3β/mTORC1/TFEB/PINK1/Parkin network reads like a conspiracy board with too much string.

The lysosome, once dismissed as a simple digestive bag, has emerged as a central player in neurodegenerative disease. TFEB sits at the control panel. And now, thanks to an AI that was slightly better at molecular matchmaking than humans doing it by hand, we have a ginkgo-derived compound that seems to know how to flip the right switches. Whether it survives the long march from bench to bedside remains the biggest unanswered question - but at least the data trail, so far, holds up under scrutiny.

References:

1. Li, A., Xiao, X., Qin, D., & Su, H. (2026). Discovery of a novel TFEB activator targeting lysosomal dysfunction in amyotrophic lateral sclerosis using artificial intelligence-based virtual screening. Autophagy. DOI: 10.1080/15548627.2026.2659295

2. Li, A., Huang, S., Cao, S.Q., et al. (2025). Isoginkgetin antagonizes ALS pathologies in its animal and patient iPSC models via PINK1-Parkin-dependent mitophagy. EMBO Molecular Medicine, 17(11), 3139-3173. DOI: 10.1038/s44321-025-00323-2

3. Wang, Y., et al. (2024). From the regulatory mechanism of TFEB to its therapeutic implications. Cell Death Discovery, 10, 1850. DOI: 10.1038/s41420-024-01850-6

4. Li, C., et al. (2022). Autophagy Dysfunction in ALS: from Transport to Protein Degradation. Journal of Molecular Neuroscience, 72, 1456-1471. DOI: 10.1007/s12031-022-02029-3

5. Settembre, C., et al. (2011). TFEB links autophagy to lysosomal biogenesis. Science, 332(6036), 1429-1433. PMID: 21617040

6. Jiang, Z., et al. (2020). Ibudilast enhances the clearance of SOD1 and TDP-43 aggregates through TFEB-mediated autophagy and lysosomal biogenesis. Biochemical and Biophysical Research Communications, 526(4), 1066-1073. PMID: 32204915

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.