#14 | Stem Cells For Spinal Cord Injuries

Quote of the Week 💬 

Stem Cell Therapies Improve Patient Spinal Cord Injuries

⌛ The Seven Second Summary: In a small Phase I clinical trial, the majority of patients who underwent stem cell therapy for their spinal cord injuries reported improvements in bodily sensations and movement. Additionally, no patient experienced a serious side effect from the treatment.

🔬 How It Was Done:

• 10 patients between the ages of 18 - 65 with spinal cord injuries from motor vehicle accidents were screened and selected to participate in the trial. 6 patients had neck injuries and 4 patients had back injuries.

• Stem cells were extracted from fatty regions in each patient’s abdomen or thigh, cultivated in a lab for 4 weeks, and subsequently reintroduced into the patient’s lower spine.

• Patients were evaluated 10 times over the course of 2 years to assess their sensory and motor skills after receiving the treatment. Their sensory and motor evaluation was on a 1 - 5 scale, ranging from complete loss of function to normal function.

🧮 Key Results:

• 5 patients demonstrated a 1 point improvement in the researchers’ evaluation scale, and 2 patients demonstrated a 2 point improvement.

  • For reference, the patients who experienced a 2 point improvement could not feel or move their bodies below the point of their spinal cord injury before this treatment. By the end of the trial, they regained some meaningful sensory and motor functions for the first time since their injuries.

• While 3 patients demonstrated no improvements in their sensory or motor function, they did not get worse.

• Perhaps most importantly, 0 patients reported serious side effects from the treatment. In fact, the most common side effects could be treated with over-the-counter medicine.

💡 Why This May Matter: Spinal cord injuries are notoriously difficult to treat because this area of the body does not grow new cells, and damaged cells do not grow new axons. This combination prevents the body from healing. However, stem cell treatments provide a way for this region of the body to receive new cells, which may lead to effective and lasting treatments that were once inconceivable decades ago.

🔎 Elements To Consider: Since this was a Phase I trial, the next step is a study with many more patients who will either receive the stem cell treatment or a placebo. It will be years before this future study concludes and the results are released.

📚 Learn More: Mayo Clinic. Nature.

Stat of the Week 📊 

An AI Agent To Solve Software Bug Issues

93 seconds

⌛ The Seven Second Summary: Researchers from Princeton University developed SWE-agent, an AI system designed to fix bugs and other software issues in code.

🔬 How It Was Done:

• The paper for this research has not been released yet, so the specifics about how the system was designed is still unclear. However, what is known is an underlying Large Language Model (LLM), such as Chat GPT-4, is used to create multiple AI agents to work together to solve a problem holistically.

• For example, there may be 3 different AI agents configured to specialize in different tasks:

  • A reasoning AI agent creates a plan to break down a problem into components and subproblems so it can be addressed in a systematic, step-by-step manner.

  • A coding AI agent writes code to answer its various subproblems in the step-by-step manner outlined for it.

  • A quality assurance AI agent runs tests to review the code it receives and provide feedback if there are errors or notable inefficiencies.

• This type of collaborative, multi-agent approach has already been shown to improve model performance beyond what a single agent can do on its own, but the specifics behind the system design is essential for generating the best results.

🧮 Key Results:

• The researchers evaluated their AI system using SWE-bench, which is a dataset of 2,294 issues and pull requests from open source Python repositories on GitHub.

• Their AI system was able to resolve 12.29% of issues autonomously, and their system had an average task completion time of just 93 seconds.

💡 Why This May Matter: Autonomous coding agents such as GPT Engineer, OpenDevin, Devika and others, are receiving a lot of attention as of late because of their rate of improvement and potential to accelerate engineering productivity as underlining models become more advanced and reliable. What is notable about this project relative to many alternatives is the speed of the its system’s task completion.

🔎 Elements To Consider: Aside from speed, code quality and costs are two other important variables to determine the practicality of these tools. With this in mind, this project still has many unanswered questions related to code quality. For example:

• What types of questions does the system struggle with and excel at? What is the difficulty level of the questions it is answering?

• How long does it take for an actual engineer to review the system’s code? Do engineers still need to make edits to the 12% of issues the system completed autonomously?

• Even if the AI system cannot solve 88% of bench problems right now, does it at least make some meaningful progress?

📚 Learn More: SWE Agent. Github.

AI x Science 🤖

A Tool To Interpret Large Vision-Language Models

Researchers from Intel and Microsoft released a paper explaining how a tool they designed can interpret and explain the rationale behind the outputs generated from a Large Vision-Language Model (LVLM).

LVLMs are AI systems with the ability to understand, reason and respond to prompts about an image or video it receives. While these systems are useful, their responses are often less reliable than AI systems that only consider text data, and text-only AI chatbots are already infamous for generating incorrect answers.

In order to improve the performance of LVLMs, engineers need to assess and refine their systems’ decision-making process. This can be a challenge, because large models use tens, if not hundreds of billions of parameters to train its predictive capabilities, and isolating the most consequential variables amongst such a vast pool of data is like finding a needle in a haystack.

To assist their explainability and fine-tuning process, the researchers developed a capability where their LVLM system generates relevancy maps and causal graphs to help them understand why the model produced a specific answer.

For example, imagine we upload an image of a happy dog. If we ask the model “is the dog happy or sad?”, a relevancy map will highlight the parts of the image the model is paying attention to when weighing variables of consideration for its answer, such as the dog’s eyes, ears and mouth. Similarly, a causal graph will outline how the model abstractly associates happiness with a dog image from its training data. This may include the animal’s facial expression, which the relevancy map already highlighted for us, as well as the presence of toys or treats, the lighting in the image, and the background environment, since these are all variables that may influence the mood of the animal.

These visual representations of the model’s decision-making allow the researchers to analyze and understand its weaknesses, and take appropriate corrective action. While their paper does not indicate specific training or performance metrics that benefited from utilizing these explainability techniques, these are the sorts of mechanisms that can significantly improve the transparency and trustworthiness of AI systems over time. arXiv.

Our Full AI Index

• Fiber Optic Cables: An international research team recently sent data at 301 TB/s, a speed 4.5 million times faster than broadband. For reference, that’s likely fast enough to send all the data on the Internet from point A to point B in less than 24 hours. Aston University. Optics Letters.

• Chemistry Queries: An international research team designed an AI system specialized in chemistry. They curated a dataset of more than 7,000 questions and found these fine-tuned models outperformed the best human chemists in their study, on average. However, their models struggled to reason through several tasks that were easy for human experts. Importantly, the models were also prone to making misleading predictions, such as in assessing a chemical's safety profiles. arXiv.

• Combating Antibiotic Resistance: Researchers from Stanford designed a new AI model to develop potential new drugs for antibiotic-resistant bacteria. The model was trained to construct potential drugs using a library of more than 130,000 molecular building blocks and a set of validated chemical reactions. During testing, the team generated 58 compounds recommended by the model, and 6 killed a resistant strain of A. baumannii. Furthermore, some of the AI-recommended compounds showed antibacterial activity against other kinds of infectious bacteria prone to antibiotic resistance, such as E. coli, Klebsiella pneumoniae and MRSA. Stanford. Nature.

• Cultural Events: The National WWII Museum in New Orleans is using AI and voice recognition technology to allow visitors to chat with World War II-era Americans through video interviews. This is a whole new way to bring the past back to life. AP News. National WWII Museum.

• Policy: The U.S. and UK AI Safety Institutes announced a partnership to test their countries’ most powerful AI models and develop a shared approach for safely developing and deploying this technology. Department of Commerce.

Other Observations 📰

A Massive Genomic Study Of Modern Birds

A team of international scientists unveiled the family tree of modern birds and determined their evolution's timing with one of the largest genomic studies of its kind.

To accomplish this feat, the researchers created a Tree of Life for modern birds by combining the genomic data from 363 bird species with nearly 200 bird fossils. Their evolutionary tree revealed most modern bird groups emerged within a short evolutionary span of one another. This time period was just 5 million years after the asteroid impact that wiped out the dinosaurs 66 million years ago.

The study also revealed a new grouping of birds that the researchers named Elementaves. This name is inspired by the world’s four ancient elements: earth, air, water and fire. This group includes birds that are successful on land, in the sky, and in water. Penguins, pelicans, swifts, hummingbirds and shorebirds represent the new group of Elementaves.

This work is the result of nearly a decade of research involving scientists from across the globe working together on the Bird 10,000 Genomes Project. The goal of this project is to sequence the complete genomes of every one of the 10,000+ living bird species. There’s a lot more work ahead for this group, but it’s always nice to see genomic progress in the field of STEM because it helps us better understand and appreciate the similarities and differences of all living species on Earth. University of Sydney. Nature.

Our Full Science Index

• Fusion Energy: Researchers from the Korea Institute of Fusion Energy announced they successfully sustained the temperature of plasma used in fusion reactors for 48 seconds at 100 million degrees Celsius. This marks a new record by the institute, and inches them closer to their long-term goal of sustaining these temperatures for 300 seconds KFE.

• Renewable Energy: Both China and Europe are experiencing record starts to the year with their respective renewable energy generation. China added over 32 GW of solar power in the first two months of 2024, an 80% year-over-year growth rate. Meanwhile, 60% of Europe's electricity was powered by renewable energy sources in January and February, driven by strong year-over-year growth in hydro, solar, and wind and a rebound in nuclear. Reuters.

• Parkinson’s: A diabetes drug, lixisenatide, demonstrated a moderate effect at delaying symptoms in patients with mild to moderate Parkinson’s disease. This was a relatively small study with only 156 patients, so while it is a promising indication, it is far from a miracle drug. NEJM.

• Life Expectancy: According to researchers from the Institute for Health Metrics and Evaluation, global life expectancy increased by 6.2 years since 1990. Southeast Asia, East Asia, and Oceania experienced the largest net gain in life expectancy at a super-regional level, totaling 8.3 years. When examining the smaller regional level, Eastern sub-Saharan Africa experienced the largest net increase in life expectancy, totaling 10.7 years. IHME. The Lancet.

• Dark Energy: Scientists analyzed the first batch of data from the Dark Energy Spectroscopic Instrument's quest to map the universe and unravel the mysteries of dark energy. A quote we loved from one of the researchers: “DESI's year-one sample of galaxies and quasars is already 6x larger than the combined measurements of all previous spectroscopic surveys conducted over the last 40 years.“ Berkeley Lab. DESI Data.

Media of the Week 📸 

Robot Folds Clothes, Sorts Objects & Answers Questions

You may recall one of our recent posts where we highlighted Figure AI’s humanoid robot washing dishes and sorting objects while talking with a human. This video is similar, though we personally think the UBTECH x Baidu integration has a faster response time to answer voice questions than the Figure AI x Open AI robot. Impressive stuff.

First Human Brain Images From A High Resolution MRI Machine

Researchers from the Atomic Energy Commission released the first MRI brain photos of its most powerful machine. It took their team 20 years to develop this technology, and the scanners of their new machine has a magnetic field intensity of 11.7 teslas. For reference, most MRI machines used in clinical practices have a magnetic field intensity of 1.5 - 3.0 teslas, meaning this machine is a ~ 4x - 8x improvement over standard machines. Specialized MRI machines are known to have intensities up to 7 teslas, yet the difference in image quality is apparent with a single glance. Atomic Energy Commission.

A 3D-Bioprinted Blood Vessel

Researchers from the European Space Agency are creating 3D bioprinted blood vessels to investigate how weightlessness changes cardiovascular systems of astronauts in orbit. They are still in the process of refining how to create these blood vessels so they are useful resources for testing. However, once complete, these 3D printed materials may be used to test the effectiveness of drugs and the treatment of vascular diseases both on Earth and in space. European Space Agency.

This Week In The Cosmos 🪐

No major events this upcoming week.

Last week we told you about the April 8^th eclipse and we hope our North American audience enjoyed the spectacle! Here is our favorite eclipse photo from Monday.