#35 | Remarkable HIV Protection

+ treating rare diseases, AI governance, a sickle cell series and more

Hello fellow curious minds!

Welcome back to another edition of The Aurorean.

We dealt with some technical issues this week that delayed our release schedule. In order to resolve the underlining problems as quick as possible, our newsletter will not go out next week. We will be back in your inbox the week of October 6th. 

With that said, wondering what STEM discovered last week?

Let’s find out.

Quote of the Week 💬

Follow Up HIV Study Once Again Shows Remarkable Protection Results

“With such remarkable outcomes across two Phase 3 studies, lenacapavir has demonstrated the potential to transform the prevention of HIV and help to end the epidemic.”

Daniel O’Day, Chairman & CEO @ Gilead

⌛ The Seven Second Summary: Another Phase III clinical trial of the antiviral drug Lenacapavir was given to a group individuals across 7 different countries, and the results were remarkable.

🔬 How It Was Done: 

  • HIV occurs from a viral infection that affects CD4 cells and weakens the body’s immune system.

  • The virus can be prevented with ~99% efficacy by taking pre-exposure prophylaxis (PrEP) drugs, which helps to block an enzyme the virus needs in order to make copies of itself and spread throughout the body.

  • Unlike common PrEP drugs, which are oral pills that require daily dosing for optimal effectiveness, Lenacapavir is an injectable medication that is administered every 6 months. The clinical trial’s goal was to compare Lenacapavir’s efficacy with an alternative daily pill.

🧮 Key Results:

  • Only 2 out of the 2,180 participants who received Lenacapavir contracted HIV, a 99.9% prevention rate.

  • By comparison, 9 out of the 1,087 participants who took Truvada contracted HIV, so Lenacapavir was 89% more effective than Truvada.

  • Similar to the results published from the previous study, these results are so convincing the trial was stopped early at the recommendation of an independent review committee, because the committee said all participants should be offered the injection because it clearly provides superior protection against HIV.

💡 Why This May Matter: The Global HIV Prevention Coalition makes it clear that despite significant progress in recent years, the world is not on track to end the AIDS epidemic by 2030 because 39 million people in 2023 have HIV. We covered the previous Lenacapavir clinical trial results a few months back, and this follow up study is more supporting evidence that Lenacapavir provides superior efficacy and convenience for patients than oral medications. If the this type of medication can be widely distributed throughout the world, then can accelerate HIV prevention efforts and help get the global response to the epidemic back on track.

🔎 Elements To Consider: Gilead charges more than $40,000 per patient per year for Lenacapavir in the United States, where it is already approved as a treatment for HIV. However, PrEP medications need to cost less than $54 a year per patient for many countries to afford them. Thus, generic alternatives need to be produced, and partnerships with Gilead need to be sorted out so financial hurdles do not restrict critical medical access around the world.

Stat of the Week 📊

FDA Approves A Pair Of Treatments For The Same Rare Disease

2

⌛ The Seven Second Summary: The United States’ Food and Drug Administration (FDA) approved two treatments for the rare genetic disorder, Niemann-Pick disease type C (NPC), within a week.

🔬 How It Was Done:

  • NPC is a progressive disease caused by changes in the NPC1 or NPC2 genes. These changes prevent the body from producing the proteins it needs to move and use cholesterol and lipids inside of cells.

  • This causes cholesterol and fatty substances to build up inside of cells in various places, which leads to organ damage and neurological issues over time.

  • The two approved treatments work in different ways. One attempts to improve the body’s protein production so cholesterol and lipids can enter and exit cells effectively to prevent harmful build ups in the first place. The other attempts to improve how damaged neuronal circuits communicate to each other, to reduce the severity of a patient’s motor and cognitive symptoms.

🧮 Key Results: Two weeks ago there were 0 approved treatments for NPC. Now, there are 2 options, each with its own promising set of clinical trial results.

💡 Why This May Matter: The average life expectancy of a patient with NPC is only 13 years old. it is now possible for a baby to be diagnosed and treated for this disease before they show any physical or cognitive symptoms. If there is an early enough intervention, the hope is these treatments will not only prolong the child’s life but also significantly improve the quality of their life. There may eventually be cases of children who grow to live relatively healthy adult lives if treatment strategies continue to improve.

🔎 Elements To Consider: 10,000+ rare diseases affect nearly 400 million people worldwide, and less than 5% of these diseases have treatments approved by the FDA. Many of these diseases cause considerable disabilities, premature death and disproportionately impacts healthcare systems, so while the NPC approvals are welcome signs, there is still a long road ahead to treat all rare diseases.

AI x Science 🤖

Credit: Google DeepMind on Unsplash

A Potential Paradigm Shift In Language Model Reasoning

Last week, we discussed the significance of OpenAI's new o1 model series. While OpenAI has been sparse with details about their model's creation, we speculate that the key advancement distinguishing this series from its predecessors is the application of reinforcement learning at the idea generation phase of a model’s problem-solving process, as OpenAI alluded to as a “natural next step” in its seminal Verify paper 2 years ago.

Reinforcement learning is not new to AI model training; it’s already embedded in various stages, such as through human feedback on final outputs. However, applying reinforcement learning to the initial idea generation step represents a fundamental shift and is arguably the most difficult step to solve.

To illustrate the point, consider that while multiple paths may lead to a correct answer, there is often an optimal route to demonstrate a deep understanding of a subject and sophisticated reasoning skills. Step 1 is inherently the point in a journey with the widest range of routes to search and consider, where a miscalculation can derail an entire plan later on. However, in spite of step 1’s significance, it is not clear how much weight to put on its responses. There will likely still be many subsequent paths a model can take to reach a relatively efficient and well-thought answer to a multi-step problem. So while incorporating reinforcement learning at step 1 has the opportunity to yield significant benefits, it is also fraught with computational complexity.

Our speculation about this specific advancement is supported by OpenAI's evaluation results. The o1 series shows marked improvement at solving deterministic problems with clear right and wrong answers, such as mathematical equations or coding challenges. However, for tasks with more subjective outcomes, like creative writing, the model's performance is the same as the previous GPT-4o series.

This discrepancy makes sense. In deterministic problems, models can be trained to recognize and generate optimal solution paths. These are the domains where reinforcement learning excels — chess, go, poker and other games provide clear and immediate feedback about performance and grade actions as good or bad fairly definitively. Math equations and code challenges are not too dissimilar from these sorts of tasks either, which is why models are continuing to improve quite rapidly in these subject matters, while their performance stagnates as the task becomes increasingly abstract.

Nevertheless, researchers are exploring alternative approaches to address this limitation, and considering the direction the o1 model is taking the industry, we think it is appropriate to cover some of these state-of-the-art techniques in the near future.

Our Full AI Index
  • Recognizing Whale Vocals With AI: Google and the United States’ National Oceanic and Atmospheric Administration (NOAA) partnered together to develop a whale bioacoustics model to identify eight distinct whale species as well as the various types of calls the species make. The model converts audio data into spectrogram images for classification, and it showed to be highly accurate across species by achieving near-perfect scores in many key performance metrics. This tool can help researchers analyze vast amounts of underwater recordings that can lead to new insights about how whales migrate and communicate to each other. Google. Frontiers. Github.

  • UN Releases Report About Global AI Governance: The United Nations' Advisory Body on Artificial Intelligence released a report outlining a blueprint for global AI governance. The report proposes seven key recommendations, such as creating an international scientific panel on AI, establishing a global fund and data framework for AI, and developing a global AI capacity network. These initiatives aim to address current gaps in AI governance, ensure equitable AI development worldwide, and foster international cooperation to help manage AI's risks and opportunities as their capabilities become more advanced. United Nations.

Other Observations 📰

Credit: Omikron/Science Source

A Series Of Articles About Gene Therapies To Treat Sickle Cell Disease

Scientific American released a great series of articles to explore the landscape of sickle cell disease: the people it affects, the pain it causes, various treatment strategies to alleviate the disease, and more. For the purposes of our write-up today, we will focus on the set of solutions that exist and are in development.

At the forefront are gene therapies to target the source of the problem: the bone marrow stem cells producing faulty red blood cells. Companies like Vertex and Bluebird have developed FDA-approved treatments using CRISPR gene editing and viral gene delivery, respectively, to either increase fetal hemoglobin production or insert resilient versions of the adult hemoglobin gene. These approaches are able to alleviate the disease’s symptoms and reduce the need for immunosuppressive drugs.

However, the high cost and limited accessibility of these cutting-edge therapies are not yet scalable solutions for patients in sub-Saharan Africa, where 75% of sickle cell cases occur worldwide. In this case, a haploidentical transplant is an alternative option, which is when patients in need receive bone marrow transplants from donors who are half-matches rather than full-matches. This approach expands the donor pool and allows more people to receive effective treatments in less time. For example, in a 54-person trial, researchers found around 90% of sickle cell patients treated with half-matched transplants were able to go at least 2 years before their bodies rejected the organ, and “some patients appeared completely cured” after receiving their transplant.

There is also a new wave of drugs to address the disease's symptoms and complications more precisely. These include antioxidants to maintain red blood cell shape, hemoglobin-binding drugs to prevent cell deformation, and treatments targeting the immune system to reduce inflammation and organ damage.

In fact, sickle cell research is part of a broader trend towards precision medicine, which we have covered on multiple occasions throughout the year. The underlying principle is to perform tests to identify a patient's unique blood and genetic biomarkers. Once this data is collected, researchers can use machine learning based solutions and other mechanisms to predict the most effective treatment strategy for a patient based on their unique physiological composition. Since the patient will receive a tailored therapeutic regimen, clinicians will be more likely to maximize a treatment’s efficacy while minimizing its unwanted side effects.

The field of medicine is still many years away from fully realizing its vision of precision medicine, but young children are already benefiting from advancements that have been made. Earlier diagnoses allow for earlier interventions that may ultimately prevent diseases from worsening as they did in the past. In fact, the NPC drug approvals we mentioned in our Stat of the Week section is an example of this development taking shape in real-time. We live in a wondrous world, and we’re thankful for all the STEM professionals and enthusiasts who help progress humanity forward. Scientific American.

Our Full Science Index
  • Another Record Year Of Solar Energy Installations: According to a new report from Ember, global solar power installations are on track to reach 593 gigawatts (GW) in 2024, a 29% increase from 2023, which surpasses most industry forecasts made at the beginning of the year. For reference, solar will likely add more GWs in 2024 than the entire global increase in coal power capacity since 2010. China continues to lead the way on this front, accounting for 56% of the total capacity. The last 2 years of unprecedented renewable energy growth now has projections for future growth aligned with the COP 28 Global Renewables and Energy Efficiency Pledge, which aims to triple renewable power capacity by 2030. Ember.

  • The Largest-Ever Observed Jets Emitted By A Black Hole: Astronomers from Caltech discovered the largest pair of black hole jets ever observed. The jets span 23 million light-years, equivalent to 140 Milky Way consecutive galaxies. The astronomers estimate the megastructure dates back to when the universe was less than half its current age, and they suspect the jets influenced how early galaxies formed. This discovery is part of a previous research effort that revealed over 10,000 similar but smaller jet systems. The amount of jet systems this project has showed indicating these structures are far more common than once believed. Caltech. Nature.

  • Physicists Observe Entangled Quarts For The First Time: Scientists at CERN's Large Hadron Collider observed quantum entanglement between a top quark and its antimatter counterpart at the highest energy levels ever recorded. Quarts are the heaviest fundamental particles known to science, and the detected this phenomenon by measuring the directions in which the quarks' decay products flew apart. The pattern of these directions revealed the quarks were entangled, with the results of the team’s measurements were so consistent that there was less than a one-in-a-million chance it happened by accident. CERN. Nature.

Media of the Week 📸

A Robotic Hovercart To Easily Move Stuff Around

Researchers from Seoul National University built a flying cargo drone. We never imagined needing a levitating cart before, but it’s exactly the type of invention that can come in handy if you need to carry something uphill or over a flight of stairs. IEEE Xplore.

A Magnetic Screen Encrypts & Displays Messages

Researchers from the University of Michigan developed a screen hat can store and display encrypted images using magnetic fields instead of electronics. The screen uses beads with orange and white hemispheres as pixels, which can be programmed by magnetic fields to display public or private images. Can already imagine how this can be used for clothing, ID badges, and other applications where traditional electronics are impractical. University of Michigan. Advanced Materials.

Striped Rock Found By Mars Perseverance Rover

Credit: NASA/JPL-Caltech/ASU

NASA's Perseverance rover recently spotted an unusual striped rock while exploring Mars' Jezero Crater rim. The rock is 20 centimeters (~8 inches) wide and its pattern is unlike anything the team has seen before on Mars. The team nicknamed the rock “Freya Castle,” and they suspect the pattern might have been formed by the planet’s ancient geology. NASA.

This Week In The Cosmos 🪐

October 2: An annular solar eclipse and a new moon. A great day to look up. It is both the best time to stargaze and a chance to see a ‘ring of fire.’

Credit: mana5280 on Unsplash

That’s all for this week! Thanks for reading.