Avi's notebook

The UK and human genetics

Sat, 18 Apr 2020 16:59:20 +0000

Science is a global practice. Most papers I read today involve collaborations, a lot of the collaborations are across countries. Many(all?) of the papers use datasets and computational methods/tools that originated across the globe. It’s sometimes interesting to take back and look at a country’s contribution to a field. I’ve worked in the field of human genetics and the contribution of the UK stands out to me.

Let’s start from the origins. Mendel wasn’t English but there’s Darwin, Wallace and Huxley. However, I’ll stick the quants for this post. One could argue that the roots of modern human genetics started in the UK. Galton introduced the idea of regression, an everyday(every hour?) word in statistical genetics as well as in science. Fisher’s contributions to statistics and population genetics will span several books. And how could we ever forget Haldane? I’d love to claim Haldane as Indian due to his later life but that would be a travesty. There’s a whole list of other population geneticists from the UK who have made seminal contributions in the 1900s.

Let’s move to technology. There’s Sanger sequencing developed by the inimitable Fred Sanger. Solexa started out in England and became today’s sequencing behemoth Illumina. Oxford Nanopore is arguably the most promising alternative to Illumina. Ion Torrent which uses current to decode nucleotides is also derived from a British company. That’s most of the sequencing companies that I know. The Sanger Institute is still one of the leading sequencing centers around the world, constantly reinventing itself as the focus of genetic studies have kept changing over the past few decades.

The role of the Sanger institute in the human genome project led by John Sulston has been well documented. Adjusted for population size/GDP the UK’s role in the project has to be the biggest. Following the human genome project, several large human genetic studies came out of the UK and have set the stage for today. The Welcome Trust Case Control Consortium was one of the first large GWAS tackling many common diseases at once. The 1000 genomes project catalogued genetic variation around the world first with low-depth sequencing followed by deep sequencing. The easy access and clear description of the data and methods from the 1000 genomes project has been unsurpassed by other genetic studies. In fact there have been much larger genetic studies in the US and elsewhere, but the data from some of these studies have not gone past a few investigators. Most of these studies result in a flagship publication for the consortium and then the data is buried in a solid state drive. The 1000 genomes project has really set the standards for sequencing studies, and sadly that benchmark is often not met.

The UK has also led the world in computational and statistical methods development for the sequencing era. Several of the most widely used aligners including BWA came out of the UK. The format specifications for files dealing with NGS also originated in the UK (SAM, BAM, VCF). This is a crucial development to Bioinformatics else we would be spending more time than we do now in the madness of converting files from one format to another. The initial round of variant callers also originated mostly in the UK. A lot of these methods formed the theoretical basis for the next generation of variant callers that sprung up elsewhere with time. Again most of the tools developed in the UK were open access, had clear documentation and well maintained mailing lists. Some of the prominent tools that came out elsewhere like the GATK were initially not open-source.

The current focus of genetic studies is on capturing diverse populations. Here too the UK set the stage and led the initial efforts. The sequencing of British Pakistanis to identify homozygous LOFs was one of the early studies. The East London project is also another promninent study in this direction.

I’ve written about biobanks and the important role they play in genetic studies in a previous post. The best established biobank today is clearly the UK biobank. The data is used by thousands of investigators around the world. This sort of data sharing is sadly an exception hardly the norm when these studies are conducted in other countries. Apart from the genotypes, the repeated phenotyping of these individuals makes this data invaluable. I would argue that the UK biobank data might be used hundreds of years down the line to look back at the state of human populations in the 2000s.

Look, I’m not a fanboy of the UK, in fact I’m pretty far from that. I’m sure we can find flaws in science from the UK just like any other country. However, the repeated excellence of UK based genetic studies particularly in terms of data sharing and open-source methods development cannot just be a coincidence. There’s something different about the priorities of the studies designed there. There is not a tendency to hide behind privacy concerns but a willingness to balance these concerns with scientific progress. The vested interests of a few investigators seems to be often set aside. I’ll drink a cup of tea to celebrate that.

I find biobanks fascinating

Sat, 07 Mar 2020 18:58:08 +0000

Human genetic association studies have been getting larger, and some of the meta-analysis use millions of individuals. Consumer genetic databases such as 23andMe are approaching(have already reached) 5-10 million consumers. These are massive studies unimaginable a few decades ago. The massive sample-sizes allow detection of variants with tiny effect. How do the variants add up to produce different phenotypes, that’s a question for the current century. We’re slowly making progress in that area. Today I want to talk about biobanks.

In my Epi class we learnt about a couple of different study designs. One common study design is the case-control study. Most of the disease GWAS use this study desing. Researchers identify a set of individuals with the desired trait as cases and another set of individuals without the trait as controls. These individuals’ genetic information is then collected and analyzed to find differences between the two groups. The phenotypic information available is whatever is available at the time of recruitment, usually a broad survey of the individual.

The second form of study design is the cohort study. Here a group of individuals is followed over time. At some point genetic information can be collected from the participants. But the key is that the individuals are followed over time. So any new phenotype or changes in phenotype can be tracked. 23andMe, Framingham heart study, Geisinger, the UK biobank are all examples of this sort of a study design. I think cohort studies are going to be super useful as we approach the next generation of genetic studies.

Right now, the Covid pandemic is on. Many individuals show mild symptoms but some individuals in various age-groups show serious symptoms. Why is this? Imagine you had healthcare and genetic data about these individuals for a long time, perhaps in say a Kaiser Permanante health system or as part of Icahn’s BioME biobank. You could start identifying risk factors, genetic or others that might correlate with disease severity. There’s no need to recruit and assemble another case-control cohort and spend millions trying to genotype them. Instead all the information is essentially already collected in the health records and can be repurposed to a new research question.

I’ve been trying to keep up with biobanks around the world. Biobanks seem to work well with nationalised healthcare systems, for example in the UK and Finland. In the US, Kaiser Permanante has a genetic research wing. Other biobanks in the US that I know of are Vanderbilt, Geisinger and BioME in NYC.

It makes sense for all large private hospitals with an interest in research to start this form of centralised information and specimen collection. This will help identify genotypic and phenotypic outliers right away.

Love Data Week talk on AI and Data Science

Fri, 14 Feb 2020 12:55:08 +0000

I attended a talk by Xiao-Li Meng yesterday. The talk was organized by the librarians here at WashU as part of a week-long celebration titled “Love Data Week”. Xiao-Li is a well known statistician(and now the founding editor-in-chief of the Harvard Data Science Review), I was really excited to attend his talk. The talk was not technical in an attempt to be accessible to a larger fraction of people. Here are my notes from the talk.

Xiao-Li started by asking the audience to ponder the definitions of artificial intelligence and data-science. The term “artificial intelligence” was first used in a Dartmough meeting in 1956. There’s a commemorative plaque (at Dartmouth?) that honors this origin. The definition of AI that stuck with me after the talk is some a machine that can mimic human intelligence. A more recent perspective seems to want AI to go beyond that, not just mimic human intelligence but augment and go beyond human intelligence. I agree these are all pretty vague ideas, I’d argue this is because we still don’t have a grasp on how to model human intelligence. The leaders in AI, according to Xiao-Li, agree that we are no way close to achieving AI at the moment, though the media would want you to believe otherwise. I recalled that this also fits with Judea Pearl’s observation from The Book of Why that current machine learning algorithms can learn patterns but can’t really go beyond that, they cannot reason for example.

Xiao-Li defined data science in terms of what it’s not. He called it the mathematician’s approach of defining something by defining it’s complement. If I remember correctly, Xiao-Li says data science is not just machine learning or statistics, data science is not just prediction, data science is not just analysis and data science is not a STEM field. He mentioned that currently several universities are racing to setup data science schools, Berkeley for example seems to have got it close to right. Xiao-Li believes data science should not be a department in itself but rather a school, just like engineering, science or humanities. At this point I’m not sure I understand what data-science is to be honest, is it some form of lightweight statistics?

Xiao-Li did a great job of talking through how we started using small samples to estimate population parameters(statistics) - starting with Bernoulli and the law of large numbers in 1713, De Moivre with the Central Limit Theorem in 1733, Anders Krier from Norway who was one of the first to propose surveys, and Morris Hansen who used sampling in the US census for the first time in the 1940s. After the US government started using surveys apparently everyone followed. Surveys were initially referred to as a “Intellectually violent revolution”.

The talk got a little technical after this(which was great, Xiao-Li breaks things down to be intuitive). He delved into why most surveys got the predictions wrong for the 2016 elections. It’s now known that there was a correlation (negative) between people who responded to surveys and people who voted for Trump. This sort of correlation is really bad for surveys, this correlation reduces the effective sample size a lot. Xiao-Li’s calculations show that even if 2.3M people might have been sampled by the surveys, this correlation reduces the effective sample size to 404 (reminds me of a 404 error). In other words the survey data is pretty close to useless. The kicker is that this correlation can only be detected in hindsight, so this could happen again. This leads to Xiao-Li’s, now popular, point - big data alone is not enough, the data needs to be representative. The size of the data can ensure a small variance but might still have bias, this can throw the survey estimate way off. I think in many ways the 2016 elections was very educational for many of us, especially the statisticians. Exit polls in many countries are still learning to optimize the methodology, I remember atleast one national election in India where the exit polls got it horribly wrong. I can predict with certainty that sites like 538 are only going to become common in every democracy.

Some of the lessons from the talk were - data quality matters nad not just quantity, don’t ignore seemingly tiny probabilistic datasets while combining datsets, look at the relative size and not just absolute size of the data. Xiao-Li also spoke about the limits of cross-validation without an independent test set. This is a fascinating topic that probably needs a lot of theory to understand properly. The gist of his argument was that cross-validation can be cherry-picked. You can do the cross-validation many times, get different performances and only report the best performance. One way to get around this is to use an independent test set that’s used just once. There’s a paper by D. Draper on the 3 CV approach that might be interesting to read. Once you have the model from cross-validation, test it on the independent data-set and see how well it does. Put your keyboards down after that, the model cannot be trained on the independent test set, it loses its value if the model is estimated using the independent test set. This is apparently what the physicists do, if the independent test set shows poor performance then the model is probably not a good one.

I had a question for Xiao-Li about the limits of prediction. In hindsight it wasn’t a well thought out question, but basically the answer depends on whether we are in the noise regime or estimating some underlying true value. If a majority of the voters continously go back and forth on their choice of candidate their final choice would be hard to predict. It was a fantastic choice of speaker by the organizers, well done!

The perfect GWAS?

Mon, 27 Jan 2020 15:13:08 +0000

GWAS come under a lot of criticism. Some of the common ones that I hear are:

Most of the effects are tiny, what do we do with this information
Most of the hits are in non-coding regions of the genome, how do these lead to a phenotype
GWAS significant hits only explain a small portion of the heritability of the trait

I wondered if there’s a GWAS that might satisfy some of these critics. I think I found one. Here’s this paper by Eimear Kenny et al. in Science 2012. The phenotype is striking and super interesting, blond hair in Solomon Islanders. The sample size of this study is 85 individuals. This population has the largest proportion of blonde hair outside Europe(5-10%). The top hit is a coding mutation in a gene called TYRP1, a melanosmal enzyme. All of the signal at this loci can be attributed to this coding polymorphism, when the effect of this polymorphism is accounted for the other hits at this locus are no-longer significant. This loci along with some simple covariates such as sex and age explains ~ 46% of the phenotypic variance.

This study is one of the earlier studies which shows how GWAS in diverse populations can uncover hits that are impossible to detect in Europeans(the predominant study population used in GWAS to date.) I find this study fascinating. The reason this GWAS is so close to perfect is due to the study design. The authors identified an interesting phenotype in an isolated population and then searched for the genetics of this trait. The trait, blond hair, makes sense for a genetic study and is unlikely to be hugely influenced by the environment, unlike a trait like say diabetes which could have a huge amount of non-genetic variance. In this case the trait had a smoking gun, a coding variant in a gene that makes sense with respect to the biology.

A lot of the genetic studies today advocate for 100s of thousands and even millions of participants. This particular study is so clever that it was able to find a hit using just 85 individuals that would not be found using millions of participants. We can argue that this phenotype is an outlier but I think there are lessons in study design here that we need to carefully think about. Sequencing everything that we can get our hands on and then figuring out why is certainly wasteful.

I have no doubt that there’s much untapped signal like this in the rest of the world. Sometimes I want to catch a flight to India and start recruiting study participants to uncover some fascinating genetics.

Mutation rate model from the Przeworski lab

Sat, 11 Jan 2020 08:08:00 +0000

I read the paper from Tomasetti and Vogelstein in 2015. As the title suggests, the paper attempts to answer an extremely interesting question of why are some cancer types more common than others. I really liked this question as unlike most groups racing to find more mutations in different cancer cohorts the authors of this paper took a step back and tried to explain the findings of the field(perhaps a bit like the Omnigenic model.) The main result of the paper was that the variation in different cancer incidences can be explained by the number of stem cell divisions in that tissue. This would mean that most of the mutations are replicative in origin and only a small fraction could be attributed to mutagens or environmental exposures. This is the “cancer is mostly bad luck” model.

The paper and the suggested finding caused a lot of furor in the cancer field. For a good read about some problems associated with this paper please see this really well written and accessible editorial by George Davey Smith et al.in the International Journal of Epidemiology here Some of the arguments against the paper were that we know from epidemiological incidences of the same cancer-type vary by region/cancer. Cancer rates also vary over time, for example drops in lung cancer and stomach cancer after relevant interventions. Does this invoke a bad-luck varies with time model. I think Vogelstein would argue from his paper that some cancers are more susceptible to environment than others. So how do we explain the correlation shown in the paper and align it with increases in mutation risk when exposed to mutagens?

I read a fascinating paper by Gao et al. from the Przeworski lab that proposed an explanation. The title of the paper is “Interpreting the Dependence of Mutation Rates on Age and Time”. I liked this paper for several reasons. Again, this paper attempted to use a simple(?) model to explain the findings to date. The model is pretty elegant and captures the core essence of the major players involved without worrying too much about the actual molecular mechanism. I find this extremely soothing to read. So how does the model work?

There are two different models. The first model is for replication dependent mutations. The model is an additive model with different mutation rates(replication error rates in this case) at different life stages. The authors split the life stages into four parts from fertilization to the production of gametes in each individual. For females the last two parts don’t count since eggs don’t replicate unlike sperm which are continously made during a male’s lifetime from spermatagonial stem cells. The total number of mutations de-novo in the offspring is then the sum of the mutations in the maternal and paternal gametes. The findings of the paper are not immediately clear. Increases in generation time will lead to stronger maternal biased mutations. One assumption is all mutations are neutral, we know this isn’t the case for example FGFR mutations in the male gametes lead to an advantage to those stem-cells. Increases in generation time lead to a stronger male mutation bias as expected. If the intercept of the extrapolated line at age zero is positive, mR,y decreases with G, consistent with the observed “generation time effect” in primates. Therefore, we argue that there is almost certainly an effect of generation time on yearly mutation rate in humans, although the magnitude of the effect could be small. we should not expect neutral substitution rates to be constant across mammalian species

The second model considers non-replicative mutations. Non-replicative mutations happen due to lesions on the DNA. Some forms of lesions on DNA on one strand cause it to not pair with the strand and this can affect DNA replication. These cells may die if the lesion is not repaired. Other lesions can pair with the wrong base pair and then replication proceeds. This would lead to one daughter cell having the right nucleotide at the position with the lesion and the daughter other cell having a mutation at that position. The model considers two processes, one a mutation process that happens at a mutation rate u and a repair process that occurs at a rate r. The number of mutations is then just a balance between these two processes. The authors consider two scenarios: 1. When the rate of repair is much slower compared to the cell division time. In this case the mutation repair is inefficient and mutations accumulate at a rate proportional to absolute time. 2. When the rate of repair is highly efficient compared to the rate of cell division. In this case, the mutation rate and the repair rate achieve an equilibrium. The cell division rate does not matter in this case. This is fascinating and suggests that no matter how fast or how quickly the cells divide the mutation rate remains constant due to the efficiency of repair. The only new mutations are the ones that are produced right before cell division and hence these haven’t had a lot of time to be repaired. These could in theory look like the replication errors, that Tomasetti and Vogelstein propose, but could actually be non-replication errors.

This simple yet elegant model produces such a fascinating result. This suggests that the correlation that Vogelstein and Tomasetti see doesn’t necessarily have to be the result of a replication error. We would see a similar pattern with mutations induced by exogenous/internal agents or if the rate of repair is highly inefficient compared to the cell division rate. So the jury is still out on this one, we cannot accept the “cancer in bad luck” model yet.

Like I mentioned earlier, I love this paper. The insights of this model will be tested soon when we begin to study mutations in more detail in different tissues, at different stages of life, different allelic backgrounds etc. What happens when the mutations confer a selective advantage or disadvantage? We are only beginning to scratch the surface at this time. I’d love to write models like these to explain and understand phenomena we see around us.

Path analysis from Sewall Wright

Fri, 06 Dec 2019 11:37:00 +0000

In 2018, when I was on vacation in Kerala, I read the Book of Why by Judea Pearl. I don’t think I’ve fully grasped the details of this book. A lot of the writing seemed to suggest that methods proposed by the authors were the right way to think about causality and I wasn’t able to evaluate this claim. I need to go back and read it again. On a higher level the book is captivating for the most part and made me wonder about what causality is and what are our best efforts as humans to-date to understand causality. The next step that we have already embarked on is to help machines learn causality.

Some of the points that I can still recall are that machine learning currently can learn patterns but cannot reason, for example current machine learning methods are not equipped to handle reasoning situations like counterfactuals(what if X happened instead of Y). Judea talks a lot in this book about Bayesian networks and why they are great to evaluate causal models. I think I’d have to spend some time reading his papers to evaluate this claim, this was not a part of my curriculum and I wasn’t able to figure out the details from the brief mathematical description in this book.

One of the points from the book that I was interested in is that Judea seems to be a great fan of Path Analysis, first introduced by the population geneticist Sewall Wright. Judea considers path analysis as the precursor to lot of the causal analysis methods currently. As a geneticist I wanted to understand why(and if) path analysis is so great and how does it work.

When I was in Don Conrad’s lab we had a journal club called the legends of population genetics. Each member of the lab was assigned one or two population geneticists and each week one person would have to present a blurb about the “legend” they were assigned. I was assigned James Crow and I’ve been a fan of his ever since. Someone else was assigned Sewall Wright. Wright is considered one of the fathers of population genetics, along with Haldane and Fisher, he played an important role in the modern synthesis of the ideas of Darwin and Mendel. Wright invented F-statistics, pioneered fitness landscapes, inbreeding co-efficients, basically a lot of the bread and butter techniques that modern population geneticists use on their shiny new data. I wanted to read up on path analysis and looked it up in my reference textbooks.

Lynch and Walsh has an appendix on Path Analysis (Appendix 2). This section is easy to digest and provides more mathematical detail on path analysis and a few applications. One of the sentences in the introductory passage of the appendix reads as follows “It is not a technique for identifying the actual sources of causality, which can only come from careful experimentation. In response to periodic abuses and criticism of this technique, Wright repeatedly emphasized this point.” I think Judea’s book missed the essence of this point, I came away from the book thinking Path Analysis could be the answer to a lot of questions related to causality. I’ll try to explain path analysis briefly below and I’m sure do a horrible job, do lookup Lynch and Walsh for a good explanation.

In Path Analysis, a path diagram is constructed. The path diagram is constructed not based on observed data but on causal assumptions on how the inputs of the system $z_i$ determine the output $y$. The path coefficients are standardized partial regression coefficients(standardizde using the standard deviations of the variables). The correlation between two nodes on the path are then specified as a product of the path coefficients and correlation coefficients between the nodes the lie on the path. Some of the segments of the path can be bi-directional which specifies that the causal relation can go both ways. Consider the phenotypes of a mom/dad/child trio as an example path diagram. The correlation between the phenotype of an individual can be written as the sum of genetic and environmental contributions. We know the direction of the path-diagram for a trio, the arrows point from the paternal genotypes to the child’s genotype and phenotype. We can thus calculate the correlation between the childs phenotype and the parent’s phenotype using the path coeffecients and correlation coefficients between the parents and the parent and child.

The path diagram is thus extremely useful in this case since the underlying causal structure is known - the direction of effects is from the parents to the child. This makes a huge difference. If we are just looking at a bunch of columns of data with no idea about the direction of effects between them the path analysis will not be of much use. Knowing a bit of prior information, the direction of causality in this case, makes a huge difference.

Wright was so far ahead of his time, there was hardly any genome sequence data when he came up with these ideas. These were all just a product of his thinking about reality.

New human genetics startup

Fri, 06 Dec 2019 11:37:00 +0000

Eight years ago if you asked me what my favorite startup in human genetics was I might have said 23andMe. Two years ago I would have said Genomics PLC.

My favorite human genetics startup right now might be Variant Bio. I found the company on Twitter the other day. The mission of the company looks super interesting, traversing the globe to find humans who are outliers for different traits. If the trait then has a genetic basis then sequencing these individuals could potentially help in finding new therapeutics that mimic the trait. It just makes sense, looking for large effect traits with a putative genetic basis. It sounds like an super exciting job for a human geneticist(definitely does to me.)

The science aside, this is a very sensitive area of research/work. Why should indigenous groups offer up their DNA data, what’s in it for them? Who benefits from this data, is it just the big drug companies or will it be shared in an open manner with the larger scientific community? Why are venture capitalists interested in funding this, wouldn’t this be better done with national funding? Are the individuals being offered genetic counseling? What happens if some unforeseen findings show up? (false paternity or incidental clinical findings for example. In some countries, adulturers can be subject to stoning.) A lot of these individuals might not have access to proper healthcare and might possibly be unaware of how genes work, any new information could potentially be stressful at the least or life-altering.

Working across cultures is extremely challenging but also I think can be very fulfilling and potentially lead to important scientific discoveries. I hope these issues are very carefully thought out and not just driven by market forces.

The omnigenic model

Sun, 24 Nov 2019 13:47:00 +0000

One of the most exciting and thought-provoking papers that I’ve read in the field of human genetics in the past five years is the Omnigenic model. There are several commentaries on the model already, for example here and here.

Once during a class, I asked a professor studying the genetics of complex traits what would be a good stopping point for his GWAS(s) or when would he consider his research question answered. I don’t think he had a good answer. The trend seemed to be collect as many samples as you can, do the GWAS, then argue that the number of samples weren’t enough to find all the loci and do more GWAS. But when will we have all the loci? The top significant hits explain a tiny fraction of the heritability. That was a bit worrying as a grad student coming in to study human and statistical genetics. It is an unwritten rule that if a GWAS finds one significant hit that uncover a new gene that could be a potential drug target the GWAS would be considered a success. We could then forget about the other significant hits or let the ‘experimentalists’ worry about what they are. I don’t know if there’s anyone following up on these other hits, almost all in noncoding regions, perhaps in the secret labs of drug companies but certainly not in the research labs that I get to hear about.

One of my favorite blog-posts makes an analogy comparing the GWAS era to the era of spotting stars in the sky without a theory of gravitation. What are all these GWAS hits? How do they function? We don’t have evidence that all of these are eQTLs, in-fact it looks like only a small fraction of them are eQTLs. What on earth are these loci? My advisor wants to know how do we study such small effects in the lab/should we even bother studying them? How do all these small effects add up to cause the trait? Is there a piece(or many pieces) of gene regulation that we don’t know about that could provide answers?

Some of the criticism against the omnigenic model is that omnigenicity is no different from polygenicity or that the idea of core genes is vague. I agree with the authors that the new term omnigenic helps with a newer conceptual understanding. There are good examples of core genes in GWAS hits that we’ve found, for example in AMD. The omnigenic model doesn’t necessarily provide all the answers, it actually makes us ask questions of the GWAS results to date. I love the fact that Jonathan Pritchard and colleagues are not just adding more loci to the GWAS pot but are taking a step back and trying to explain the findings that we have so far. It just makes sense to do this. Are gene networks the answer? How can we study gene networks with our expanded genome engineering toolkit? Questions abound as the genetics revolution marches on. As a computational geneticist I sometimes just can’t stop thinking about this problem. I’d love to chat over coffee or beer if anyone’s interested in talking more about this in person.

Genetics seminar series: Fall 2019

Sat, 16 Nov 2019 18:37:00 +0000

The genetics department at Washington University hosts a seminar every Thursday. The tradition of inviting outstanding scientists to speak goes back a long time(I have no idea when this started). As graduate students we get a chance to eat lunch and interact with the speakers from other institutions. I wanted to write briefly about some of the seminars I’ve attended in Fall 2019.

Rick Lifton gave the keynote for the Precision Medicine pathway retreat. Dr. Lifton had a bunch of interesting human genetics cases that he spoke about. Some of the cases Dr. Lifton described had parallels to X-men, except for the fact that his patients are real humans. One of them had an extremely low fracture rate due to his outlier bone density(many z-scores out). The story is described here. Another case that he talked about was someone who had to constantly drink pickle juice to maintain appropriate salt levels. This research is described here . The genetic causes of his condition helped shed light on genes involved in and mechanisms of hypertension. This strategy of identifying outlier individuals for some phenotype, checking if the trait is observed in several members of the individual’s family and then narrowing down on the genetics behind the trait is now a strategy adopted by several human genetics groups. The UK BioBank and similar biobanks elsewher(Japan, Finland, Estonia and Vanderbilt come mind first) seems well positioned to identify such cases. Other resources are actively being setup, I learnt about a similar resource being developed at Stanford currently with data from wearables being actively collected. In discussions with my advisor it’s a bit strange that more of the medical institutions in the US haven’t raced to do this with the EHR data available. This is happening behind the scenes right now and the countries with centralized health systems seem to be ahead of the rest at the moment.

Hunter Frasier from Stanford gave what I found to be one of the most intriguing seminars of this semester. Dr. Frasier’s group and collaborators used an approach called CRISPEY to precisely induce single nucleotide changes in yeast strains. This is described in this paper This approach could be helpful to identify the causal nucleotide change in a QTL. QTLs are usually loci that encompass a lot of genetic variants and identifying the causal variants is still an open problem. Dr. Frasier’s group tested variants present in extant population. The results were super interesting, most of the changes with larger effects were non-coding SNPs instead of coding SNPs. One way this result could make sense is if the large effect coding SNPs had already been weeded out by natural selection in extant populations, this makes sense. However the strange part was the causal QTNs showed little overlap with the loci detected by traditional crossing methods. This is strange and potentially very interesting.

There were several other seminars that I was able to attend and found interesting. Eimear Kenny gave a super interesting talk on the population genetics of NYC. The stories from this cohort were very interesting including a SNP predicted to have a huge effect on height(I think its 6 - 10 inches) They also developed newer faster methods to detect IBD in large cohorts. Some of the results from the talk are in this paper. This is one of the most diverse cohorts in the United States at the moment and I imagine working with these samples must be super exciting. The lunch with Dr. Kenny was super interesting as the students touched on different issues including genetic privacy. At a later seminar, Stephen Montgomery spoke about the state of co-localisation approaches for GWAS variants. Most variants tested in GTEx look like eQTLs in at-least one tissue. So testing if a GWAS variant is also an eQTL is not a simple lookup and can lead to a false-positives. Instead sophisticated statistical methods which test if the structure of the signal in the GWAS haplotype matches the structure seen in the eQTL signal are now the norm. However the problem linking a GWAS hit to a gene still still isn’t solved. Dr. Montgomery also spoke about identifying rare variants in gene expression outliers and how this information could potentially be used to improve PRS predictions. This reminded me again of the human genetics approach of identifying “all the outliers”. Jennifer Phillips-Cremins gave a super interesting seminar on how 3-D genome structure is altered at the breakpoint of repeat expansion disorder patients. This change in the TAD structure could potentially alter transcriptional regulation or could be the result of such an alteration. Dr. Phillips’ group has some amazing technology for creating loops between loci, this could be used to examine the downstream effects of loop formation. Sabrina Spencer gave a fascinating about non-genetic hetrogeneity and tracing down why all cells don’t cycle through the cell-cycle in an identical manner. Dr. Spencer’s group has developed a live-cell sensor to monitor CDK2 levels and used this to identify the cells which don’t progress through the cell cycle. Some of these are found to be cells with likely DNA damage. This area of research is pretty fascinating and could potentially uncover a lot of important cell biology.

I found the Fall 2019 edition of the Genetics seminar series to be interesting. The department seminars could potentially be an important part of graduate student training. I hope the tradition continues in the following years.

Updates

Sun, 29 Sep 2019 13:10:00 +0000

Like most bloggers I’ve stopped updating the blog after a while. This post is not a promise to pick up the pace of this blog! I thought I’d just share some anecdotes from life.

My daily life now involves a bit of cell-culture, reading, seminar, meetings and less coding. While I do miss coding and statistical thinking I’ve picked up some experimental skills which give me a different perspective on science. There are definitely non-linearities in this line of work. Most of the time progress is small but sometimes these can add up to some major progress.

I was tutoring a student for ACT-prep and there was a question on number of bacterial colonies surviving His- media. I felt at ease tackling it having worked with yeast for a couple of months.

I got hold of a book called “Bioinformatics The Machine Learning Approach”. One of the authors is someone who worked on neural nets to predict MHC binding to small peptides. This was one of the really early applications of neural networks to this problem and is possibly still one of the best predictors currently after more than a decade or two. I was curious to check out the book but I’m still in the middle of reading it.

I’m TAing an Introduction to Biostats class. The class teaches introductory Python and some basics of statistical thinking. Teaching programming has always been fun to me and I’ve done this with several colleagues ranging from informal help with the command-line/programs to a 7-day course for IT professionals in Chennai. I think I could do 10-20% teaching and 80-90% research in an ideal job.

Being in grad school has meant that I get to grab lunch with the seminar speakers. I’ve been lucky to have lunch with many fabulous scientists. I recently learnt that one of the speakers that I had lunch with was knighted (Sir Peter Donnelly)! While such titles don’t mean anything to me it’s still a bragging right.

Being in graduate school also means that I get free dinner twice a week at school, breakfast once a week and sometimes lunch once a week. While the food might not be at the same level as that at Google, DropBox etc, it definitely is a perk and means students end up working longer on the nights with free dinner!

I’ve been reading a lot of books since grad school. Two recent books that I read were Animal Farm and Things Fall Apart. I wish both had been a part of the CBSE syllabus in India. I guess there’s only so many books you can require students to learn.

Talking about high-school, the Math and Computer Science that I learnt in high school have helped me immensely. Kudos to my high school CS teacher Uma Arti for two great years of C++. The curriculum was at such a level that one could find a job at Google with this course and some additional effort. The high-school CS syllabus could include a bit more about data structures and Algorithms and this would be sufficient. Perhaps it already does this, it’s been 14 years since I graduated. Again, there’s only so much room in the syllabus. But hands-down learning programming early has been immensely helpful to me. I remember learning C during a summer camp in sixth grade, most of the other attendees were adults but the instructor was happy to take me on in the class. Sometimes all you need is someone to take a chance on you. Indeed this has been a major factor in my life.