El llenguatge de la vida

 

L'Eric Topol ens ofereix un resum d'alta qualitat per entendre el moment que viu la ciència mitjançant la intel·ligència artificial. 

Impressionant. Aquesta és la llista de grans models de llenguatge en ciències de la vida (LLLMs):

1. Evo. This model was trained with 2.7 million evolutionary diverse organisms (prokaryotes—without a nucleus, and bacteriophages) representing 300 billion nucleotides to serve as a foundation model (with 7 billion parameters) for DNA language, predicting function of DNA, essentiality of a gene, impact of variants, and DNA sequence or function, and CRISPR-Cas prediction. It’s multimodal, cutting across protein-RNA and protein-DNA design.

   Figure below from accompanying perspective by Christina Theodoris.

2. Human Cell Atlas A collection of publications from this herculean effort involving 3,000 scientists, 1,700 institutions, and 100 countries, mapping 62 million cells (on the way to 1 billion), with 20 new papers that can be found here. We have about 37 trillion cells in our body and until fairly recently it was thought there were about 200 cell types. That was way off—-now we know there are over 5,000.

   One of the foundation models built is Single-Cell (SC) SCimilarity, which acts as a nearest neighbor analysis for identifying a cell type, and includes perturbation markers for cells (Figure below). Other foundation models used in this initiative are scGPT, GeneFormeR, SC Foundation, and universal cell embedding. Collectively, this effort has been called th “Periodic Table of Cells” or a Wikipedia for cells and is fully open-source. Among so many new publications, a couple of notable outputs from the blitz of new reports include the finding of cancer-like (aneuploid) changes in 3% of normal breast tissue, representing clones of rare cells and metaplasia of gut tissue in people with inflammatory bowel disease.

3. BOLTZ-1 This is a fully open-source model akin to AlphaFold 3, with similar state-of-the-art performance, for democratizing protein-molecular interactions as summarized above (for AlphaFold 3). Unlike AlphaFold 3 which is only available to the research community, this foundation model is open to all. It also has some tweaks incorporated beyond AlphaFold 3, as noted in the preprint.

4. RhoFold For accurate 3D RNA structure prediction, pre-trained on almost 24 million RNA sequences, superior to all existing models (as shown below for one example).

5. EVOLVEPro A large language protein model combined with a regression model for genome editing, antibody binding and many more applications for evolving proteins, all representing a jump forward for the field of A.I. guided protein engineering.

6. PocketGen A model dedicated to defining the atomic structure of protein regions for their ligand interactions, surpassing all previous models for this purpose.

   

7. MassiveFold A version of AlphaFold that does predictions in parallel, enabling a marked reduction of computing time from several months to hours

8. RhoDesign From the same team that produced RhoFold, but this model is for efficient design of RNA aptamers that can be used for diagnostics or as a drug therapy.

9. MethylGPT Built upon scGPT architecture, trained on over 225,000 samples, it captures and can reconstruct almost 50,000 relevant methylation CpG sites which help in predicting diseases and gauging the impact of interventions (see graphic below).

   

10. CpGPT Trained on more than 100,000 samples, it is the optimal model to date fo predicting biological (epigenetic) age, imputing missing data, and understanding biology of methylation patterns.

11. PIONEER A deep learning pipeline dedicated to the protein-protein interactome, identifying almost 600 protein-protein interactions (PPIs) from 11,000 exome sequencing across 33 types of cancer, leading to the capability of prediction which PPIs are associated with survival. (This was published 24 October, the only one not in November on the list!)

Al KBR, Cartier-Bresson

I per aquí aprop, mentrestant anem venent privadament la recerca inicial finançada públicament i perdem l'oremus, mentre alguns se n'aprofiten. Desgavell perfectament dissenyat.

IA pertot arreu

Multimodal biomedical AI

En aquest blog he anat explicant darrerament els avenços en la intel·ligència artificial i els podeu trobar en aquestes entrades. La innovació és tan ràpida que costa molt estar atent al que està passant. Per aquest motiu us suggereixo una ullada a l'article que parla de la intel·ligència artificial multimodal en biomedicina, recullo una frase:

The development of multimodal AI models that incorporate data across modalities—including biosensors, genetic, epigenetic, proteomic, microbiome, metabolomic, imaging, text, clinical, social determinants and environmental data—is poised to partially bridge this gap and enable broad applications that include individualized medicine, integrated, real-time pandemic surveillance, digital clinical trials and virtual health coaches (Fig. 1). In this Review, we explore the opportunities for such multimodal datasets in healthcare; we then discuss the key challenges and promising strategies for overcoming these. 

I un gràfic: 

 

i el que són els foundation models:

Bommasani R et al, arXiv, July 2022

i si voleu saber l'última informació sobre "multimodal ai", la trobareu aquí.

I jo em pregunto, hi ha algú aquí aprop que estigui a l'aguaït de tot això? Es tant important que convindria no perdre passada i em temo que fem tard.

PD. Avui  a WSJ i al NYT 

"This is going to change everything about how we do everything. I think that it represents mankind’s greatest invention to date"

Searching for a healthy ageing

The Biology of Inequalities in Health

The Lifepath research consortium aimed to investigate the effects of socioeconomic inequalities on the biology of healthy aging. The main research questions included the impact of inequalities on health, the role of behavioral and other risk factors, the underlying biological mechanisms, the efficacy of selected policies, and the general implications of our findings for theories and policies. 

 The impact of socioeconomic condition on premature aging is mediated by known behavioral and clinical factors and intermediate molecular pathways that Lifepath studies have revealed, including epigenetic clocks (age acceleration), inflammation, allostatic load, and metabolic pathways—highlighting the biological imprint (embodiment) of social variables and strengthening causal attribution.

 There is still a wide gap between social and natural sciences, both on methodological and conceptual grounds. Natural sciences focus in particular on biological mechanisms and outcomes, i.e., they address “zoe” (biological life), while social sciences address “bios” (biographical life), if we refer to the terminology used by Ronald Dworkin. In fact, epidemiologists aim to connect zoe and bios in meaningful ways, though this attempt has rarely become explicit. An exception is the work of Nancy Krieger who proposed the concept of “embodiment.” Biology and biography (124) meet in the health status of an individual, depending on social position at a given age. These concepts start to be incorporated into epidemiological research, via the integration of social contexts and biomarkers in a life-course approach. The results from analyses carried out within Lifepath suggest that the socioeconomic environment, from early life and across the life-course, is an important risk factor for health and exerts its effects via intermediate biological mechanisms.

Great research!

PS. Austin Frakt in NYT Putting a Dollar Value on Life? Governments Already Do

Edward Hopper

Lamarck returns

Lamarck's Revenge: How Epigenetics Is Revolutionizing Our Understanding of Evolution's Past and Present

I've just started to read this amazing book. Chapter 1 says:

Charles Darwin espoused evolution as driven by natural selection. However, an earlier theory, proposed more than a half century before the first publication of Darwin’s greatest work, came from a naturalist whose life and work were limned by the flames of the French Revolution.

Lamarck arrived at a three-step process in what was to be the first really rational explanation for what we now call “organic evolution.” First, an animal experienced a radical change of the environment around it. Second, the initial response to the environmental change was some new kind of behavior by that animal (or whole species). Third, the behavioral change was followed by morphological changes that were heritable in subsequent generations. This proposed process came to be named after its author. Today, a variant on what Lamarck proposed is sometimes called “neo-Lamarckism,” but more often “epigenetics,” or “heritable epigenetics.”

Jean-Baptiste-Pierre-Antoine de Monet, Chevalier de Lamarck, had a different view about heredity and why animals changed through time. His scientific beliefs were that things that happen to us during our lives can change what we pass on to our next generation, and perhaps into even further generations. Darwin knew well what Lamarck theorized. Darwin believed that his own theories about evolution could not coexist with any aspect of what Lamarck postulated. We now know this is no longer the case.
Lamarck’s Revenge looks anew at what are, perhaps, humanity’s most basic questions: the “where,” “when,” and “why” of getting to the present-day biota on this planet. But the vehicle to do this is by asking specifically about the “how.” What were the evolutionary mechanisms, the balance between Darwinian and neo-Lamarckian (aka heritable epigenetics), that produced not only our physical biology but some aspects of our heritable behavior as well?

Here are some possibilities. First, that the process known as epigenetics combined with periods of extraordinary environmental change has played a far greater role in what is called the “history of life” than is accepted by all but a small cadre of revolutionary biologists. This is perhaps most decisively shown through the epigenetic process of “lateral gene transfer,” where on a given day, in a given minute, some organism is invaded by another and a product of that invasion is the incorporation of vast numbers of new genes, making the invaded creature something else again, neither the invader nor the invaded. This is known.
Second, new evidence points to a probable role of epigenetics in producing rapid species transitions by mechanisms other than lateral gene transfer. Science has discovered that major evolutionary change of a species can happen a thousand times faster by epigenetics than by the process demanded by the Darwinian theory of single, random mutations along a creature’s genome or DNA (or, in some cases, RNA). This is most likely to occur during and immediately after rare, major environmental perturbations (such as mass extinctions and their aftermath).

A controversial view of epigenetic inheritance

A critical view on transgenerational epigenetic inheritance in humans

A new article in Nature suggest that more evidence is needed to ascertain the role of epigenetic inheritance.

Even if the molecular mechanisms exist to transmit epigenetic information across generations in humans, it is very likely that the transgenerational transmission of culture by communication, imitation, teaching and learning surpasses the effects of epigenetic inheritance and our ability to detect this phenomenon. Cultural inheritance has certainly had an adaptive role in the evolution of our species, but the evidence for transgenerational epigenetic inheritance, as laid out above, is not (yet) conclusive. 

Let's wait for new evidence...

Clinical utility of genomic sequencing

The Path to Routine Genomic Screening in Health Care

Now that whole genome sequencing is knocking at the door of the clinician, it is the time to ask for clinical utility. The understanding of how such information will change diagnostic and therapy is paramount. There is still no need for cost-effectiveness, clinical utility comes first.
And the editorial at Annals explains exactly this issue, highly recommended:

There should be little doubt that individually tailored health care management plans based on DNA analysis are coming, but the timing of their introduction into routine clinical care is contingent on further demonstrations of clinical utility and proven implementation models.

My impression: let's wait for epigenetic biomarkers, beyond whole genome sequencing that provides less than 100 actionable genes out of 20.000. Though,

 The fact that only a small percentage of people would benefit from GS today is counterbalanced by growing evidence that the benefit could be significant, and perhaps even life saving

 

Pepe Castellanos at Barnadas Gallery

The meta-informational challenge of molecular data

The future of DNA sequencing

Where does DNA sequencing goes from here?. Nowadays, this is an appropriate question to pose.  The answer appears in an article in an interesting article in Nature.

Now, geneticists would like to have DNA sequences for everyone on Earth, and from every cell in every tissue at every developmental stage (including epigenetic modifications), in health and in disease. They would also like to get comprehensive gene-expression patterns by sequencing the complementary DNA copies of messenger RNA molecules.

In a mere 40 years, the central goal of putting molecular data about cells to practical use has changed from an informational challenge to a meta-informational one. Take clinical applications of genome-sequence data. It may soon be possible to use DNA sequencing routinely to analyse body fluids obtained for any clinical purpose. But only a vast amount of well-organized data about the multi-year medical histories of millions of people will provide the meta-information needed to establish when to ignore such data and when to act on them.

Integrating genome and epigenome studies

The Key Role of Epigenetics in Human Disease Prevention and Mitigation

I've said it many times: beware of snake-oil sellers. Nowadays you may find it everywhere, specially on internet. You may get a genetic test for a disease that creates a false illusion of safety, or another that provides an unnecessary and avoidable concern. Only evidence based prescribed tests can be considered appropriate.
Therefore, if you want to confirm that genome is not enough, you have to check the review at NEJM on epigenetics. At the end of the article you'll find the explanation on why we do need integrated genome and epigenome association studies. You'll understand that cancer is fundamentally an epigenetic disease.
The current knowledge is changing quickly some conventional truths and "known unknowns" that we've had for years. This is good news for citizens, and bad news for snake-oil sellers if detected. Governments should help citizens on this screening effort, and protect citizens from fake medical information.

Approaching the golden age of epigenomics and epitranscriptomics

A new twist on epigenetics

If epigenomics is crucial to discard the genetic predestination paradigm, now we can add a new 'omics to the paradigm: epitranscriptomics. Last February, Nature published interesting news related to recent scientific developments:

The epigenome helps to explain how cells with identical DNA can develop into the multitude of specialized types that make up different tissues. The marks help cells in the heart, for example, maintain their identity and not turn into neurons or fat cells. Misplaced epigenetic marks are often found in cancerous cells.

 Chuan He and Tao Pan are two researchers that have been working on new ways of controlling gene expression

He and others have shown that a methyl group attached to adenine, one of the four bases in RNA, has crucial roles in cell differentiation, and may contribute to cancer, obesity and more. In 2015, He’s lab and two other teams uncovered the same chemical mark on adenine bases in DNA (methyl marks had previously been found only on cytosine), suggesting that the epigenome may be even richer than previously imagined.

The team had shown for the first time that RNA methylation was reversible, just like the marks found on DNA and histones.

Methylated adenine bases are the focus of research on gene expression.

Genome editing, closer than you think

Human Genome Editing Science, Ethics, and Governance

Last week the US patent office ruled that hotly disputed patents on the CRISPR revolutionary genome-editing technology belong to the Broad Institute of Harvard and MIT. In a former post I explained the dispute. Genome editing in my opinion shouldn't be patented and will see exactly the impact of such ruling in US and elsewhere in the next future.
If you want to know in detail what does genome editing means for the future of life sciences, have a look at NASEM book.

It is now possible to insert or delete single nucleotides,interrupt a gene or genetic element, make a single-stranded break in DNA, modify a nucleotide, or make epigenetic changes to gene expression. In the realm of biomedicine, genome editing could be used for three broad purposes: for basic research, for somatic interventions, and for germline interventions.

CRISPR (which stands for clustered regularly interspaced short palindromic repeats) refers to short, repeated segments of DNA originally discovered in bacteria. These segments provided the foundation for the development of a system that combines short RNA sequences paired with Cas9 (CRISPR associated protein 9, an RNA-directed nuclease), or with similar nucleases, and can readily be programmed to edit specific segments of DNA. The CRISPR/Cas9 genome-editing system offers several advantages over previous strategies for making changes to the genome and has been at the center of much discussion concerning how genome editing could be applied to promote human health.

I would just want to say that these patents destroy the soul of science, since access should be available with no barriers for the development of  innovation. Patents are not the incentive for discovery in this case, as I explained in my post, natural processes should'nt be patented. And this is why today is a really sad day.

PS. My posts against patents

Michael Kiwanuka. Home again