chrisprobert

insitro | Machine Learning for Drug Discovery | South San Francisco, CA | Full Time | Onsite

insitro is reinventing drug discovery by bringing cutting-edge machine learning in a closed loop with our high throughput robotic biology data factory.

Current software roles include:

- Machine Learning Engineer

- Data Engineer

- Head of Data Engineering

See: http://insitro.com/jobs or feel free to email me: [email protected]

I'm curious whether this is backed by Google File System (https://static.googleusercontent.com/media/research.google.c...), or something else.

What's interesting about this approach (TIL expansion) relative to other leading cellular immunotherapy approaches (CAR-T/NK and TCR) is that it doesn't rely on gene editing. Long term, I think genetically "programmable" cellular immunotherapies are more likely to win (e.g. because they can be programmed to overcome tumor immune suppression), but it's impressive that a durable response is achieved here with clonally expanded TILs.

There is a great review on progress in cell based immunotherapies here: https://www.cell.com/cell/abstract/S0092-8674(17)30064-8

Coupa Cafe on Ramona St. in Palo Alto, or the one in Y2E2 at Stanford.

Totally agree!

Announcing TensorFlow's new development roadmap mandate: copy everything PyTorch is doing :-)

Great to see this coming from a team with such a solid AI background!

I would put > 50/50 odds on there being a human alive today with at least one CRISPR-edited germline variant. I think the question is now how can we regulate/control CRISPR germline editing; not how can we prevent it.

Human embryo gene editing was reported in May 2015 (http://dx.doi.org/10.1007/s13238-015-0153-5). It's reasonable to assume that the editing took place significantly before the submission date, especially given reports that the paper was first rejected from several other journals on ethical grounds (http://www.nature.com/news/chinese-scientists-genetically-mo...). I'm not trying to suggest that these particular scientists have performed experiments on viable embryos also, but I'd be very surprised if someone hasn't.

Key question for digital health / biotech startups is whether direct to consumer (D2C) advertising for testing services (e.g. genetic carrier screening; cancer risk screening) would be included.

Correct - their current v4 chip covers ~602K SNPs and they have > 1 M customers, so you couldn't uniquely identify all of them by genotype. But given the frequency of rare variation in humans, you'd expect to be able to reduce a given genotype to a much smaller set of customers.

You're right that Ancestry.com is focused on mtDNA, but 23andMe actually uses a custom Illumina microarray with autosome, allosome, and mitochondrial targets. Here's a study where 23andMe used their (old version) chips to replicate a bunch of known genetic associations, most of which are autosomal: http://journals.plos.org/plosgenetics/article?id=10.1371/jou...

Totally agree here - would be easy to fabricate DNA oligos with given mtDNA or STR sequences, especially if you know forensic investigators are only going to use very small target sequences for identification purposes.

This could be a case for using shotgun / whole genome sequencing - we'd expect to see a more even distribution of genome coverage than one would get by leaving targeted oligos behind. But this isn't likely to happen anytime soon; the costs are far too high (10-100X greater than targeted sequencing).

In the meantime though, one highly feasible avenue for spotting synthetic DNA oligo fragments is the presence and position of nucleosomes (https://en.wikipedia.org/wiki/Nucleosome), which are DNA-associated protein complexes that occur in DNA from organisms, but not in synthetic DNA oligos. These have been used successfully to trace tissue of origin in cell free DNA in humans, and have specific signals related to chromatin/genome topological state. The only way I can think of to fake this signal would be to have a cell culture from the person you're trying to imitate. Granted, this too is not completely unreasonable - there are now very robust commercially available protocols for deriving iPSCs from small dermal fibroblast samples.

Good question. This problem affects multiple steps of a NGS-based forensics product: sample collection, DNA extraction, library preparation, the sequencing itself, alignment/assembly, and statistical variant interpretation all have potential for large biases/error modes that could affect the specificity of these types of methods.

One specific answer is that the FBI is gearing up to regulate new devices in this area (https://www.fbi.gov/about-us/lab/biometric-analysis/codis/st...). For example, in this publication (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3757157/), they specifically analyze an IonTorrent PGM for use in forensics applications, which we'll probably see them do for various other platforms that come on to the market.

Separately, and outside the forensics realm, there's a trend towards increased regulation of DNA sequencing. For example, NIST has developed/is developing methods to evaluate sequencing platforms: http://www.nist.gov/mml/bbd/dna-022514.cfm. This is relevant to other sequencing applications too (e.g. personalized medicine, somatic tumor profiling, etc). The FDA are also involved here, but more focused on medical applications.

So, I think collectively through both the increase in the forensics community regulating forensics NGS applications, and more broadly the biomedical science/technology community regulating general NGS platforms, we'll see good technology validation standards (at least in the U.S.). But the significantly higher complexity of these systems does introduce more opportunity for error, so it's entirely possible we'll see similar biases in NGS based forensics.

This article brings up problems with current PCR-based STR genotyping methods, but lacks information about technologies that are competing to be the future of forensic genotyping. Here's some more context on the NGS technologies that are set to replace it:

Illumina is pursuing mitochondrial DNA (mtDNA) sequencing. An advantage here is that cells contain many copies of mtDNA sequences (as opposed to just one copy of each gDNA haplotype), and mtDNA contains hyper-variable regions which confer strong individual specificity. This is potentially advantageous in crime scene samples, where the DNA could be damaged through degradation processes like sun exposure. http://www.illumina.com/areas-of-interest/forensic-genomics/...

Ion Torrent/ThermoFisher are going after the same STR targets, but using their Torrent and Proton NGS platforms (rather than PCR). Unlike regular PCR methods, this can provide things like allele frequency estimates, and can call more than one base into variable regions (which provides more information, and can potentially be used to infer things like height, ethnicity, hair or eye color). https://www.thermofisher.com/us/en/home/industrial/human-ide...

Carlos Bustamante (a Stanford Professor, and world expert in ancient genomics / diverse population genomics: https://med.stanford.edu/profiles/carlos-bustamante) has founded IdentifyGenomics, which is a startup focused on new methods for forensic DNA sequencing (disclosure: I know Carlos, but I'm not involved in his startup).

Definitely an important problem, and will be interesting to who succeeds in converting forensic investigators to use NGS at scale.

I also enjoyed his post on the "Two Cultures of Computing": http://pgbovine.net/two-cultures-of-computing.htm

Definitely something worth thinking about as we build tools for other developers/engineers to use.