I'm the creator of Origami Assays and am happy to answer any questions.
Screening for COVID19 is an urgent problem, but the infrastructure for running these assays is limited. While there are important efforts underway to make more tests available, one simple and low cost way to help is to use the assay infrastructure we already have more efficiently.
The idea is that rather than run 1 assay on 1 patient sample, we intelligently pool patients samples and run our limited assays on these pooled samples. It is analogous to data compression, but for assays instead of files.
Nonadaptive pooling designs are well studied branch of applied mathematics and engineering, and are well suited for COVID19 population screening for the following reasons:
(1) Binary assay
(2) Low positive rate
(3) Large number of samples
These three features mean that the data stream coming from COVID19 assays is nicely compressible.
I've put together a series of examples of nonadaptive pooling designs for COVID19 that I'm calling "Origami Assays". These designs provide a few things:
* Concrete examples, with performance metrics for a range of sizes of designs.
* Software infrastructure for decoding designs with error estimates.
* A low cost, DIY paper template system for constructing complex pooling design mixtures by hand.
Advantages:
* Yields up to an 11.9x improvement in patient testing throughput (for the XL3 assay design).
* Can be rolled out immediately, on any existing assay platform (RT-PCR, antibody, LAMP, or qSANGER)
* Pool design can be done by hand without extensive training.
Disadvantages:
* Pooled designs can call false positives if too many positives are present in the population.
* Pooling can dilute samples
* Constructing pooling designs is a mind numbing task for humans.
I'm trying to roll out Origami Assays to all who may benefit from them. Any questions, thoughts, or ideas are welcome!
Isn't the low positive rate kind of an assumption though, because of currently limited testing? If this is much more widespread than we thought, will this mechanism fall over?
The low positive rate is a constraint, but in practice we see that population screens are yielding between 0.5% and 4% positive rate, depending on the sampling population/scenario.
There are many use cases where we expect a low positive rate too. For example, an employer screening what appear to be healthy employees.
A nice thing about these designs is that if they get overloaded, they call false positives and the decoder can indicate when the design limits are exceeded.
In this case, we would need to do a second round of testing for validation--often on a small handful of cases.
Yes, much like a Bloom filter, but instead of 2, it gives 3 output types:
1) not in set
2) possibly in set
3) in set
Depending on the input (sample population), it is possible to get results with all 3 states. Origami Assay's decoder differentiates the "(3) in set" from "(2) possibly in set" for efficient post-testing.
And yes, multiplex designs are used in quite a few lab settings. Quality control in factory settings and all over the place in bioinformatics.
And yes, you can overcome the false positives by retesting. The nice thing about nonadaptive designs is that in most cases you don't need to re-test because the right answer just falls out. If you do need to retest, the design flags that too. Yay for math!
The article says that this strategy is best fit for when most of the patients are negative. I'd like to see how the number of false-positive change given a change in the number of patients that test negative (increase and decrease).
I'd also be interested in a comparison of how reliable it is compared with the current test strategy.
Reliability is an interesting question. For the monoplex case, we put all of our eggs in one basket, so if we miss it once we miss it forever. In the multiplex scheme, we have the possibility of recovery because we build in multiple tests.
Isn't the dilution of the virus a problem if most of the patients are negative? You were using a test on a sample containing 100% virus positive blood, but now you are using a test on a sample that contains x% virus positive blood, with x potentially small
True, dilution is a challenge but the assays used for COVID19 detection generally have a pretty wide dynamic range (3-5 orders of magnitude). The multiplex designs in Origami Assays mix between 4 (for S3) and 36 (for XL3) samples per well, depending on design. This means that samples are diluted by a max of 1:36 in the largest XL3 design (2.8% original concentration) so the assay used must be able to handle that.
Alternatively, samples may be concentrated depending on the specific assay.
11 comments
[ 4.2 ms ] story [ 33.6 ms ] threadScreening for COVID19 is an urgent problem, but the infrastructure for running these assays is limited. While there are important efforts underway to make more tests available, one simple and low cost way to help is to use the assay infrastructure we already have more efficiently.
The idea is that rather than run 1 assay on 1 patient sample, we intelligently pool patients samples and run our limited assays on these pooled samples. It is analogous to data compression, but for assays instead of files.
Nonadaptive pooling designs are well studied branch of applied mathematics and engineering, and are well suited for COVID19 population screening for the following reasons:
(1) Binary assay (2) Low positive rate (3) Large number of samples
These three features mean that the data stream coming from COVID19 assays is nicely compressible.
I've put together a series of examples of nonadaptive pooling designs for COVID19 that I'm calling "Origami Assays". These designs provide a few things:
* Concrete examples, with performance metrics for a range of sizes of designs. * Software infrastructure for decoding designs with error estimates. * A low cost, DIY paper template system for constructing complex pooling design mixtures by hand.
Advantages:
* Yields up to an 11.9x improvement in patient testing throughput (for the XL3 assay design). * Can be rolled out immediately, on any existing assay platform (RT-PCR, antibody, LAMP, or qSANGER) * Pool design can be done by hand without extensive training.
Disadvantages:
* Pooled designs can call false positives if too many positives are present in the population. * Pooling can dilute samples * Constructing pooling designs is a mind numbing task for humans.
I'm trying to roll out Origami Assays to all who may benefit from them. Any questions, thoughts, or ideas are welcome!
There are many use cases where we expect a low positive rate too. For example, an employer screening what appear to be healthy employees.
A nice thing about these designs is that if they get overloaded, they call false positives and the decoder can indicate when the design limits are exceeded.
In this case, we would need to do a second round of testing for validation--often on a small handful of cases.
1) not in set 2) possibly in set 3) in set
Depending on the input (sample population), it is possible to get results with all 3 states. Origami Assay's decoder differentiates the "(3) in set" from "(2) possibly in set" for efficient post-testing.
Question for you -- are these types of methods used in labs already?
It seems the false positive issue could be overcome if you just re-test anything that is positive separately, wouldn't that work?
And yes, multiplex designs are used in quite a few lab settings. Quality control in factory settings and all over the place in bioinformatics.
And yes, you can overcome the false positives by retesting. The nice thing about nonadaptive designs is that in most cases you don't need to re-test because the right answer just falls out. If you do need to retest, the design flags that too. Yay for math!
I'd also be interested in a comparison of how reliable it is compared with the current test strategy.
https://www.smarterbetter.design/origamiassays/default/encod...
Reliability is an interesting question. For the monoplex case, we put all of our eggs in one basket, so if we miss it once we miss it forever. In the multiplex scheme, we have the possibility of recovery because we build in multiple tests.
Alternatively, samples may be concentrated depending on the specific assay.