Blogroll

Controlling humidity levels in your home is extremely important. Its presence can contribute to wellness challenges, as well as affect the integrity of your home and possessions.

At a glance

• We report on the recent publication of our retrosynthesis model RetroChimera in the journal Nature (opens in new tab).
• The paper describes the model’s architecture as well as extensive validation studies, including the model’s ability to recall rare reaction types, and successful zero-shot transfer and fine-tuning on proprietary datasets.
• We open-source RetroChimera’s implementation and weights in the hope that it will enable researchers to accelerate development of new medicinally relevant molecules and advanced materials. 

Developing new medicines and materials requires making new molecules but planning how to make them is still largely manual, time-consuming, and costly. RetroChimera automatically proposes high-quality synthesis routes. The model combines two strong models with complementary strengths, learning how to rank their proposals to produce better predictions than either alone. In blind tests, PhD-level chemists prefer RetroChimera’s individual reaction predictions over preceding models and recorded literature reactions. 

Custom-made molecules are unlocking advances in modern medicine, smart materials, and sustainable agriculture. Yet, progress is slowed by chemical synthesis—the time-consuming process of making new molecules from simpler building blocks in the lab. In addition, synthesis is a significant driver of drug development costs. So even as computational methods make it possible to explore large numbers of novel molecules, finding practical ways to synthesize them remains a critical challenge. 

Figure 1: Planning a synthesis by working backward. Retrosynthesis starts with a target molecule and proposes successive disconnections into simpler precursors until purchasable building blocks are reached. The highlighted path shows a complete synthesis route; pale branches illustrate alternatives explored along the way. Circles represent molecules and squares represent reactions. For clarity, only a few branches are illustrated, with chemical structures shown for the target, one intermediate, and selected building blocks. 

Retrosynthesis approaches this problem by working backwards from a target molecule, breaking it down step by step into simpler precursors (Figure 1). This process is comparable to playing strategic board games like chess and Go. It involves contemplating a wide range of possible immediate moves, or individual disconnections, while also requiring high-level strategic thinking to reach the end-to-end synthesis plan. However, the number of possible moves in retrosynthesis is much larger than in board games, and it is not obvious which moves would be available for a given molecule. Existing systems face major challenges, including recalling rare but strategically important reactions, robustness beyond the training distribution, and aligning with chemists’ expectations. As a result, retrosynthesis often requires highly specialized expertise, which hinders scaling and automation of scientific discovery.

Figure 2: Our framework for ensemble-based retrosynthesis with learned re-ranking which underpins RetroChimera. The ensemble receives a target molecule as the input, which is then processed by the sub-models. The model outputs are aggregated using a learning-to-rank strategy. While in this work we only investigate deep learning models as prediction sources (solid boxes), it is possible to add additional sources, for example calls to reaction databases or human-in-the-loop queries (dashed box). 

In a paper recently published in the journal Nature (opens in new tab), we present RetroChimera (opens in new tab), a new framework for retrosynthesis prediction. It is built around two models (Figure 2). R-SMILES 2, a Transformer-based de-novo model, predicts precursor molecules directly from the input molecule. This gives it the flexibility to learn reaction patterns directly from data. However, its unconstrained generation can also make it prone to hallucination.

NeuralLoc, in contrast, is a graph neural network- (GNN) based model that encodes both the target molecule and reaction templates as graphs. It selects reaction templates and predicts where they should be applied to the target molecule. Its predictions are grounded in reaction patterns extracted from the training data, so it tends to produce more accurate and reliable outputs. But it’s more constrained when encountering reactions not covered by the template library. 

These differences actually turn out to be a strength. Rather than making the same kinds of predictions, the two models capture complementary patterns in chemistry and specialize in different reaction types. R-SMILES 2 performs particularly well on reactions that involve large changes over the course of the reaction, while NeuralLoc excels in reactions of low precedence and those involving more localized changes.

RetroChimera combines the ranked predictions of both sub-models using a learned ensembling strategy. Each model assigns a learned, rank-dependent vote to each predicted reactant set, and votes are added when both models propose the same reaction. By learning how much to trust each model at different ranks, RetroChimera can leverage their complementary strengths, approximately matching the better-performing sub-model across reaction classes.

Figure 3: Expert assessment of multistep synthesis routes. Left: Ratings of individual reaction steps. Right: Complete routes accepted or rejected for ten challenging targets. RetroChimera succeeded on nine targets, versus five for the de novo model, four for the editing model, and two for NeuralSym, a strong baseline model. 

As a result, RetroChimera performs strongly across both common and rare reaction classes and produces retrosynthesis predictions that better align with chemists’ judgment (Figure 3). In blind tests, expert chemists preferred disconnections of complex molecules suggested by RetroChimera over those obtained from its constituent sub-models, as well as those from more established approaches, and even from the test set itself.

We believe RetroChimera could help researchers identify promising synthesis route more efficiently, supporting faster design-make-test cycle across molecular science applications, including drug discovery and design of smart materials. RetroChimera could enable chemists to assess more—and more complex—candidate molecules at large scale. Paired with increasing levels of laboratory automation, we expect further acceleration toward closed-loop, self-improving systems for synthesis planning and execution.  

RetroChimera is available on GitHub (opens in new tab) (MIT license) and accessible via Microsoft Foundry (opens in new tab). For instructions on how to access the checkpoint, we refer to the GitHub repository.

We invite the broader chemistry community to experiment with RetroChimera, helping us identify its strengths and shortcomings so we can enhance it in the future. We are looking forward to hearing how it performs on various targets you care about!

For a deeper look at the findings, including evaluation experiments with our external research collaborators, see the full Nature publication (opens in new tab) and the accompanying Microsoft Source article (opens in new tab). 

Publication OSS repository Microsoft Foundry Opens in a new tab

The post Improving synthesis prediction of small molecules at scale with RetroChimera appeared first on Microsoft Research.

When you hear Kubuntu, you might think of the Linux-based operating system by the same name. But have you heard of Kubuntu Focus? It's a separate hardware-focused brand that sells machines with Kubuntu pre-installed and pre-configured. I've spent the last few weeks with the Kubuntu Focus Ar Gen 1 laptop and I have to say, I'm not all that impressed with the hardware.

The streamers are going all out for the last full week of September with a terrific lineup of new movies. First up is a music biopic about a landmark folk singer before he went electric. If you like rom-coms, you'll appreciate Prime Video's selection this week.

VS Code is one of the most popular editors out there—it is relatively lightweight, has an extension for everything, and has a core that is open-source. However, the actual download you get from Microsoft isn't; that version comes with telemetry, a proprietary license, and a marketplace that is limited to Microsoft's own builds.

Alexa and Google's Gemini are convenient, but they have trade-offs. You pay for convenience with your privacy, subscriptions, and sometimes even ads. Beyond the obvious privacy concerns, there is also a practical performance issue: most of the processing done by those types of devices takes place on cloud servers, which can introduce noticeable latency.

Sound matters. Among the many labels you’ll encounter when setting up a soundbar or home theater, “ARC” and “eARC” are sure to pop up. But what do they actually mean, and why should you care?

When you get a new phone or device, the first thing people think about is what to do with their current one. More often than not, they trade it in to get money back or lower monthly payments on their new device.

Months after the initial tease, the first Googlebooks are available to pre-order today. The Android laptops are as high-end as promised and can compare well against rival Windows PCs — at least, if you're heavily invested in the Google ecosystem.

TL;DR: Dunkin' is giving away 1 million free drinks on Sept. 22. The promo code will drop at some point during the day, so keep this page bookmarked for an update.

Dunkin' is giving away 1 million free drinks, again. That seems like an absolutely wild promotion, but Dunkin' actually does this kind of thing fairly regularly. That doesn't make it any less exciting, and the process of claiming remains the same.

Dunkin' doesn't release specifics on what's actually included in the giveaway until the promo code goes live. That code will drop sometime on Sept. 22, after which point everyone will rush to claim one of the free drinks. We'll be keeping a keen eye on all the usual channels to bring you that code as soon as it goes live. These freebies go quickly, so you need to be ready.

SEE ALSO: Starbucks just dropped a Snoopy cup for fall, and yes, I'm taking the bait

Once the code goes live, you'll need to enter it in the "Offers & Rewards" section of your app. Yes, that does mean you need to be a Dunkin' Rewards member, but it's free to sign up, and this type of giveaway comes around pretty often. It's worth the hassle.

SAVE $50: As of Sept. 21, the Samsung 75-Inch M70H Mini LED 4K Smart TV is on sale at its lowest-ever price of $547.99 at Amazon. That's $50 off its list price of $597.99.

Opens in a new window Credit: Samsung Samsung 75-Inch Class M70H Series TV $547.99 at Amazon
$597.99 Save $50   Get Deal

With big sale events like Prime Big Deal Days right around the corner, it's the perfect time to start shopping around for a new TV. If you don't want to wait until then to buy, Amazon already has plenty of great deals on select models ahead of its next sale.

Among them is the Samsung 75-Inch M70H Mini LED 4K Smart TV, which has dropped to $547.99 for a limited time.

This is a $50 discount off its list price of $597.99, but more importantly, it's the lowest price this TV has ever hit. For a 2026 model, that's a very nice offer to take advantage of before October Prime Day starts.

SEE ALSO: How to sign up for Amazon Prime in time for the next Prime Day

With 4K resolution, you'll enjoy sharp details in your favorite movies, shows, or games on this Samsung TV, while Mini LEDs deliver bright, vibrant colors and deep contrast. Plus, the 120Hz refresh rate ensures fast-paced action on screen looks smooth and sharp, which is a nice bonus for gamers or people who enjoy watching live sports and don't want too much motion blur.

On top of all that, it's also a smart TV. If you're someone who's signed up to a variety of different streaming apps, you can access them all easily from the home screen to jump into whatever you're watching.

There's no telling how long this deal will last, though. If it's caught your eye, act fast to save on the Samsung 75-Inch M70H Mini LED 4K Smart TV at Amazon.

The best deals this week, hand-picked by our shopping experts

• Apple iPhone 18 Pro — free at T-Mobile 

• Paramount+ — $0.99 for 2 months with code p2mcs26prem

• Apple AirPods Pro 3 — $199 $249 (save $50)

• Apple iPad 11-inch — $399 $449 (save $50)

• TCL 85-Inch Class QM8L Series TV — $2,497.99 $3,999.99 (save $1,502)

America is obsessed with true crime. I once talked to my friend's dad, who told me he puts on a documentary every night and falls asleep to "death and crime." It's a bit morbid, but it doesn't take away from the fact that these scandalous docuseries are extremely popular. You can find true crime on Hulu and HBO Max, but Netflix is still one of the top destinations for true crime.

John Oliver's main target was American health care conglomerate UnitedHealth Group and its insurance arm, UnitedHealthcare, during Last Week Tonight on Sunday.

But the host also found time to get in a quick dig at David Ellison, the CEO of Paramount Skydance, who is currently in the process of trying to acquire Warner Bros. Discovery. If the deal closes, Ellison's company would also own WBD subsidiary HBO, which airs Oliver's show.

"Having access to wealthy parents just cannot be the system to get badly-needed healthcare, even if it is apparently the system that lets you be in charge of HBO," says Oliver, bringing up a photo of Ellison alongside his billionaire father, Larry Ellison, who personally pledged $40 billion to support Paramount's bid.

"Yeah, my soon-to-be business daddy has a literal business daddy, and to be very clear, I deeply respect that. And I would never make fun of you for it, unlike that pesky Stephen Colbert.

Want more of the best of late night? Sign up for Mashable's Top Stories newsletter.

Take a look inside Bluecore Energy, a company newly out of stealth, building a prototype nuclear power plant on a barge.

Learn more about the future of nuclear and marine energy: Inside the world's biggest bet on fusion energy
The orb-like Ocean-2 turns wave power into renewable energy

Timecode chapters
0:00 Intro
0:29 Inside Bluecore energy
1:07 Bluecore’s plan
1:40 The reactor
2:05 Water desalination
2:41 Command center
3:12 Bluecore Energy’s big goal

Video credits
Host: Jesse Orrall
Scripted by: Jesse Orrall
Producer: Jesse Orrall
Camera: Dustin Shepard
Video editor: Jesse Orrall
Thumbnail designer: Viva Tung

My Plex library had grown far too large, and it was time to downsize it. I could delete stuff, sure, but I wanted to keep most of what was there. So, I turned to converting my entire library to H.265, and it saved me nearly 10TB of space.

TL;DR: To celebrate Pokémon's 30th anniversary, The Pokémon Company released a new card set with high pull rates to stop scalpers. A factory red-ink glitch created "Blood Mew" — an accidental, creepy variant of Mew #151.

To celebrate the franchise's 30th anniversary — and cool down tension around short stock and scalper buyouts — Nintendo and The Pokémon Company dropped a commemorative TCG set loaded with higher pull rates for shiny variants and fan-favorite Pikachus. The goal was noble enough: make cool, rare cards more accessible for casual fans and active players alike.

Naturally, that didn't stop the trading circuit from finding a brand-new, ultra-exclusive prize to lose their minds over.

Enter "Blood Mew."

Originally designed as a serene, golden illustration of card number 151, a batch of the new Mew cards suffered a heavy printing defect at the factory. Likely caused by a foil splice error during manufacturing, Mew appears to be floating in — or descending into — a thick pool of deep red ink saturated across half the card instead of hovering over a bright magical grove.

Card trader Miss Natoshi spotted the factory mishap, posted it online, and gave the sinister-looking misprint its instant-classic nickname.

This Tweet is currently unavailable. It might be loading or has been removed.

In the Pokémon card world, factory mistakes are treated like gold simply because they're so rare. While standard shiny cards are printed by the millions, an accidental ink leak or misaligned foil creates a one-of-a-kind card that can't easily be remade.

Even though Nintendo tried to make this 30th Anniversary set easy for everyone to buy, the internet hype machine has already turned "Blood Mew" into an expensive viral circus.

SEE ALSO: The Pokémon TCG 30th Celebration Greninja ex Box gets a price cut on Amazon — act fast to save $10

Hoping to hit the jackpot, resellers are listing their misprinted cards on auction sites and social media to see how much collectors will pay. Some super optimistic sellers are even listing their cards with asking prices in the six figures.

Whether the hype stays this high or cools off, one thing is for sure — a simple printer mistake just accidentally created the most Metal card in Pokémon history.

Pentalobe screwdrivers up! iFixit's teardown of the iPhone 18 Pro lets you peek inside one of Apple's latest signature devices.

In order to evaluate how repairable iPhone 18 Pro and Pro Max are, the outlet's lead teardown technician, Shahram Mokhtari, gives a close-up look at the mechanical details — from the camera's variable aperture to the Face ID infrared camera newly hidden beneath the display.

Mokhtari opens up the back panel and finds most repairs will still need to go through the screen, a tricky (and expensive) process involving managing adhesive and the plastic screen frame. And you'll get a good look at Apple's next-generation C2 cellular modem and A20 Pro-powered logic board.

For more on the iPhone 18 Pro Max, Mashable's Stan Schroeder has reviewed the device. Or perhaps you're looking for a hands-on with the iPhone Duo?

Want to learn more about getting the best out of your tech? Sign up for Mashable's newsletters or get Mashable push alerts.

I recently went through my box of old computer junk and discovered an old laptop. After checking it out, I discovered it had 8GB of RAM, a slow quad-core processor, and a battery that is ready to quit after 10 minutes.

Buying a 10GbE switch for a network where almost everything still runs at 1GbE sounds unnecessary. If your computers, access points, printers, and other devices still have Gigabit Ethernet, spending money on a 10GbE switch can look like buying hardware for a problem you do not have. That was my situation, and I still bought a 10GbE switch.

King Charles, an AI version of Princess Diana, and Ed Sheeran are the unusual combination brought together in SNL UK's latest cold open, during which a D.I.A.N.A AI-bot goes rogue and the king is forced to beg for Sheeran's help as the apocalypse nears.

"I couldn't help but overhear your conversation. You see I was just outside, sitting on the fence," says Sheeran (Annabel Marlow), who has faced criticism this week after Macklemore was dropped from his tour for pro-Palestinian statements. "You know sometimes, when it comes to conflicts, the bravest thing you can do is to say nothing at all."

Cue an impressively cringe-inducing song about "both sides" being "equally right".

Elsewhere, the singer was also roasted in the show's "Weekend Update" segment.

Want more film and TV news and reviews delivered straight to your inbox? Sign up for Mashable’s Top Stories newsletter or get Mashable push alerts.