Your immune system, your sleep, your AI — all linked to depression

Depression may be an immune system problem as much as a brain problem

What if the most important thing happening in severe depression isn't in your thoughts — it's in your blood?

A research team led by Michael Maes, a psychiatrist who has spent decades studying the biology of severe depression, has published a sprawling theoretical framework pulling together 36 years of research from labs in Poland, Germany, Bulgaria, and the US. The core argument: in acute, inpatient-level depression, your body's immune army — specifically Th1 cells (a type of immune cell that activates inflammatory responses) and M1 macrophages (immune cells that, when switched on, drive inflammation) — goes into high-alert mode. Imagine your kitchen smoke alarm going off not because of a fire but because you made toast. The alarm is real, the response is real, but it's misfiring. That misfiring, Maes and colleagues argue, cascades outward. It disrupts how your body handles iron. It lowers good cholesterol. It drains your supply of omega-3 fatty acids. It affects tryptophan — an amino acid your brain converts into serotonin. And it drives oxidative stress, which is the biological equivalent of rust building up inside cells. The team proposes calling this whole chain NIMETOX: neuroimmune–metabolic–oxidative dysregulation. The evidence is striking in places: T-cell activation patterns from inpatient MDD cohorts could distinguish depressed patients from controls with 91% specificity. Supervised machine learning identified inpatient depression as a qualitatively distinct biological group, not just a point on a spectrum. The catch, and it's a significant one: this is a narrative review, not a clinical trial. Maes draws heavily on his own lab's cohorts. There is no systematic search protocol, no meta-analysis. What this paper offers is a map — a coherent story about how these pieces might fit together — not proof that they do. Whether treating the immune system can relieve depression is a separate question that remains wide open.

Scientists gave AI agents a 'dial' for anxiety, depression, and PTSD — and watched what happened

What if you could turn anxiety up with a dial, watch an AI freeze in its tracks, and then turn it back down to zero?

A team of researchers built software agents — reinforcement learning agents (programs that learn by trial and error, receiving rewards for good actions and penalties for bad ones) — and added seven parameters they call 'appraisal knobs.' Each knob shapes how the agent interprets threats and rewards. Turning one up is a bit like slowly closing the vents in a house: the agent starts to avoid more, freezes more, and stops exploring. The researchers call this the anxiety phenotype. Other knobs produce what looks like mania, addiction, PTSD, obsessive checking, impulsivity, and depression — seven disorders in total, each testable across more than 1,000 experimental runs. The results are specific in an important way: each knob produces a smooth, dose-dependent change in behaviour, and the effects don't bleed into each other. The anxiety knob doesn't accidentally create mania as a side effect. When knobs were switched back off, most disorders reversed — but anxiety and PTSD did not. Those required a gradual 'exposure-with-response-prevention' curriculum to recover. That detail mirrors something real: exposure therapy is indeed the standard treatment for both in humans. The disorders also self-organised into a two-dimensional map — reward-seeking on one axis, threat-avoidance on the other — which the researchers didn't build in deliberately. It emerged. The catch: these are software agents in grid worlds, not brains. Calling their behaviours 'depression' or 'PTSD' is a useful model, not a literal claim. Whether a 2D grid-world experiment tells us anything reliable about a human mind is a very long road. What the framework does offer is a cheap sandbox for generating and testing predictions — and that is genuinely useful.

One number from your sleep and movement habits may flag depression risk

How much of your depression risk is already hiding in the rhythm of your days — before you feel anything at all?

A research group analysing the CHARLS longitudinal survey — data from 15,233 older Chinese adults — asked whether all the messy information in your sleep habits, physical activity, napping patterns, and social timing could be compressed into a single number without losing much predictive power. They called it the Circadian Rhythm Score, or CRS. Think of it like a credit score, but for your body clock: dozens of separate behaviours folded into one figure. Their machine learning model, built on gradient-boosted trees, achieved an AUC of up to 0.83 in some age groups — meaning that using only the CRS, the model correctly ranked who was and wasn't at depression risk about 83% of the time. (AUC, or Area Under the Curve, is a measure of how well a model separates two groups; 0.5 is chance, 1.0 is perfect.) Beyond the score itself, the team found specific thresholds worth noting. Roughly 300 MET-minutes of exercise per week appeared to be a minimum effective dose for depression risk reduction — that is approximately 60 minutes of moderate-paced walking, five days a week. A nap of around 65 minutes was protective for sleep-deprived individuals. A nocturnal sleep window of about 6 hours emerged as a sweet spot, with both shorter and longer windows associated with higher risk. The catch: this is an observational study of older Chinese adults. Correlation, not causation — we cannot say that hitting these targets will prevent depression. The sample is also culturally and demographically specific, and whether the CRS would generalise to other populations is untested. Treat the specific numbers as hypotheses, not prescriptions.

Today's three papers are each, in their own way, asking the same question: what is depression, really, and where can you grab hold of it? The NIMETOX framework says look at the body — the immune system and metabolism are co-conspirators, not bystanders. The RL agents paper says build a working model of how cognitive appraisal generates disorder-like behaviour, so we can run experiments cheaply and at scale. The circadian rhythm paper says look at the daily rhythms hiding in plain sight — sleep, movement, timing — for signals that appear before a clinical crisis. What connects all three is a move away from treating depression as a single, monolithic thing. Each paper implicitly argues that depression is a system — biological, computational, behavioural — and that you have to understand the whole system before you can reliably intervene. That is a harder task than finding one drug or one biomarker. But the map is getting richer, and today added three meaningful pieces to it.

The NIMETOX framework will be worth following if clinical trials targeting immune pathways in treatment-resistant depression — several are already underway, including anti-inflammatory drug trials in MDD — start reporting results later this year. For the RL agents work, the critical next move is whether the framework's quantitative predictions hold up against real patient data, not just grid worlds — watch for follow-up preprints from the same group. And if the Circadian Rhythm Score gets validated in a non-Chinese, non-elderly cohort, that would be a meaningful step toward real-world deployment in wearable health tracking.

• Wikipedia: Neuroinflammation and depression background
• Wikipedia: Computational psychiatry
• Wikipedia: Circadian rhythm