The Four Common Problems in Multiplex Panel Design

Hello, everyone. Welcome to our webinar, “The Four Most Common
Problems in Multiplex Panel Design.” My name is Ted Dacko and I’m going to serve
as your host today, and the webinar will be presented by John SantaLucia. Hi, everyone. Here is our agenda. I’d like to introduce the speaker in just
a second. Then I would like to tell you a little about
our sponsor today, DNA Software. Then we will cover the four problems of Multiplex,
we’ll provide a solution for the four problems of Multiplex, and actually provide you with
a demonstration of what a solution might look like. Then we’ll talk about the benefits. Then we’ll to a quick summary for you, and
then we will answer your questions and we’ll talk about next steps. I’d like to introduce our speaker, John SantaLucia. John has been a professor at Wayne State University
for over 23 years. He is a world renowned biophysical chemist
with over 50 papers and 1 book. He served on numerous government panels and
he’s been a consultant to numerous companies as well, many of which are on the phone today. And I do need to point out that we have well
over 100 people registered for this webinar, so this is a very interesting topic. He measured DNA nearest-neighbor thermodynamic
parameters and he is the co-inventor of qPCR CopyCount. He founded DNA software in 2000. For the first ten years, he was the Chief
Science Officer, and for the past seven years he has been the President and CEO. John, welcome. Thank you, Ted. John, my first question to you is, can you
tell the audience a little bit about DNA Software and why you founded the company? Sure, Ted. Back when I was a professor, we were studying
the fundamental aspects of DNA folding and DNA hybridization, and mismatch thermodynamics,
and I started getting a lot of phone calls from different companies and academic labs
asking if I could consult with them to help solve their problems. It became obvious to me early on that there
was a commercial need to solve the problems that people had and I couldn’t do it in my
academic lab. So we started DNA Software to try to meet
those needs. That’s all I wanted to say about DNA Software
for now. We’ll get to share more as we go throughout
the seminar today. Can you tell us what these four problems of
Multiplex PCR are? Sure. First, let’s discuss the applications of
Multiplex PCR and then I’ll share with you how that relates to the problems. The first application of Multiplex PCR is
for detection of infectious disease. A lot of folks want to design panels of tests
that will detect more than one pathogen at a time, such as for example, upper respiratory
pathogens or a group of messenger RNA found in human blood, et cetera. The key there is to be able to account for
target variation and that’s called consensus design. There you want to be able to distinguish target
of interest from near-neighbors and background sequences. Another application of Multiplex PCR is for
target enrichment. Here you might have something like 100 genes
that are cancer causing genes or involved in pathways involving cancer. You want to enrich those targets for Next-Generation
Sequencing. That type of application involves very high
levels of multiplexing, often from 100-plex to higher levels of amplification needed. There the trick is to get even amplification
of all of the targets. And lastly, there is the need to do strain
genotyping, particularly for SNPs or determining the strain of viruses and bacteria, for example. There’s a need to not only detect genomes
of different viruses and bacteria, but also to distinguish them. To summarize this, designing Multiplex PCR
is hard, and we want to show you that we appreciate the nuances of that field. Here are the four most common problems, to
answer your question, Ted. The first is the problem of false negatives. That is a test where a person has a particular
disease, but the test is negative despite the fact that they have the disease. Those are caused by a variety of things that
we’ll be going over. False positives occur when a person does not
have a disease or the pathogen DNA is not present in the test tube, but the test says
that it is. That’s a false positive. We’ll be showing why that arises and what
the solutions are to that problem. The third problem is the problem of coverage. This has to do with if you have a lot of different
variants of a target of interest or if you have many different targets that you’re trying
to Multiplex. How do you know that you can detect all of
those different targets at the same time? The first three topics really are technical
in nature. The last topic about company resource constraints
is more physical. It has to do with … Well, we’ll cover that
when we get to it. Which of these four, in your mind, is the
biggest challenge for organizations? Well actually, the last one. The company resource constraints is the largest
problem, because often times people don’t realize what’s possible and don’t realize
that they have a need that they help. Okay, well let’s get into it. The first topic is about false negatives and
that has to do with sensitivity of assays. What are the major causes of false negatives? It’s widely known this first cause, “1A”,
is that target secondary structure can inhibit primer binding. That’s particularly true for RNA targets,
but often true as well for DNA targets. We’ll cover that topic in detail. In addition to that, there is a number of
other causes of false negatives. Those can be everything from false amplification
due to primer dimers, false amplicons, primer-amplicon interactions, and unimolecular extension. All three of these last causes, B, C, and
D, are due to polymerase extension which causes the depletion of primers and NTPs. There’s one other cause of false negatives
as well, and that’s sequence variation. We’ll cover that topic in problem #3 when
we talk about coverage. Let’s first talk about target secondary structure. Most users, when they are considering primer
hybridization, they use a two-state model and they may even use some of the work that
we published many years ago now, like the nearest neighbor model, to predict the difference
in energy between the random coil state and the duplex state. It turns out that that model just is not sufficient
to predict what’s going on. The reason is that real DNA is not straight
like this when it’s in a random coil. In fact, real DNA molecules are folded. Shown on the right-hand side is a more sophisticated
model, an N-state model, many states are included here. Here’s a hybridization region in green and
a two-state model would say, here’s a hybridization region and here’s a primer, and they want
to hybridize as shown on the right-hand side here. But in fact, there’s competing equilibria. The target itself can be folded and in fact,
that region to which you want to direct your primer can be folded and that folding inhibits
the hybridization. You need to pay the energy cost to break that
folding before a primer can bind. Also, the primer itself can form hairpin interactions,
for example. Such hairpin interactions can sometimes be
innocuous, and other times that hairpin interaction can cause other problems if that hairpin can
serve as a template for a polymerase (more later about this). As I mentioned here, the problem with the
two-state approach, is that there’s an energetic cost to break secondary structure. That folding is one of the major causes of
low sensitivity in single-plex PCR and definitely in multiplex PCR, where it’s much more complex. In particular with multiplex PCR, such folding
of DNA can cause the amplification to be uneven. That is, one amplicon amplifies faster than
other amplicons. One of the major causes of that is that some
of the targets are more folded than others. If you haven’t simulated the folding of your
target DNA, there would be no way for you to know that. The solution to this problem is to solve for
the amount bound, to solve the coupled equilibria. We have ways at DNA Software to predict all
of these equilibria and solve for the amount that is in the duplex state, and that amount
is what is ultimately related to your assay sensitivity. ) Let’s now cover a little bit about the other
causes of false negatives. One of them, I mentioned, is the idea of formation
of primer dimers (Figure B). Just by accident, it can happen that two primers
have pairing at their 3′-prime end and a polymerase would extend those. That is problematic for PCR and multiplex
in particular, because it (i.e. polymerase extension) depletes the PCR primers. They get chemically used up and the NTPs get
depleted as well, and that can ultimately cause the multiplex reaction to fail. Another type of interaction, which is particularly
important in multiplexing, is primer-amplicon interactions (Figure C). Shown here in this picture (Figure C) on the
upper right side is a target from the Zika virus, but there’s a primer binding site for
one of the other primers that are in your multiplex to detect other targets. Let’s say there are additional primers to
detect Influenza A. Unfortunately, this Influenza A primer cross-hybridizes to the Zika target. That cross-hybridization reaction can completely
ruin that amplicon by making it shorter and making it so the probe doesn’t bind, and therefore
you get a false negative. So those are two really important things to
account for, and I think the first one (i.e. primer dimers), most primer design software
account for. The second one is more subtle and harder to
do. This last one is unimolecular extension (Figure
D) and I have a question for you to think about. I’ve shown here five different primers. The sequences are similar to each other but
they’re all a little bit different, and I have this question: “Which of these structures
do you think is extensible?” It’s not obvious at first blush, and the true
answer is: “it depends”. In fact, all of them are extensible in one
way or another and I’ll show you what I mean by that. Here are some considerations (bottom right). First of all, some of the primers end with
a mismatch at the end, and some mismatches are extensible and other are not. For example, on this first case on the left-hand
side, there’s an A-A mismatch. A-A mismatches are generally not very extensible. Even though this is a thermodynamically stable
hairpin, this particular hairpin is unlikely to be problematic for PCR. On the other hand, this primer in the middle
differs by one nucleotide. Just change to a C at the end, and now it’s
an A-C mismatch. Many polymerases are tolerant to A-C mismatches
and they will extend them. Once they extend that primer to the end, that
primer has now changed its chemical composition and that primer will no longer be viable in
the PCR reaction. Recall that I said that almost all of these
could in fact be substrates for polymerase extension. That would happen if you happen to choose
a polymerase that has 3′-exonuclease activity. If your enzyme has 3’-exonuclease activity,
it will chew back and remove some of these dangling-end nucleotides until it creates
a species that is in fact extensible. That’s very problematic, and we generally
don’t recommend enzymes that have 3′-exonuclease for multiplex applications. Lastly, perhaps the most surprising of all
is the structure that’s on the right-hand side. The right most structure has its 5’-end
base paired. We wouldn’t normally think that such a structure
would extensible by a polymerase, but think about what happens during a PCR reaction. During PCR, the complement of that sequence
would be made, and once that complement was made, the complement would be highly likely
to fold back on itself and form a structure with base pairs at the 3’end of the antisense
strand, and such structures cause the formation of amplicon dimers or higher order concatemers,
which also are problematic. They will shut down the PCR completely. This brings us to the second type of problem
in PCR, which is the problem of false positives. False positives are the result of false hybridization
occurring in your reaction. For multiplex PCR, designing your primers
to be specific is absolutely critical, because in a multiplex PCR there are so many primers
and so many amplicons that the probability of cross-hybridization becomes huge and grows
very quickly. In fact, it grows exponentially. I will show you in a moment that the tool
that most people use for such specificity tests is a BLAST search, but it’s really not
the best tool for the job. The second thing about false positives is
the idea in part C, that the multiplex is a complex interacting system. False amplicons can involve rare hybridization
reactions and weak hybridization reactions that involve mismatches or even bulges in
them. A tool like BLAST just is not up to that task. What we need is a tool that can scan huge
numbers of primers against large numbers of genome databases, and that is a challenging
problem. Let’s take a minute and talk a little bit
about an approach that many scientists take, which is to use the program BLAST, to predict
their primer specificity. Of course, I don’t need to explain much about
this to this audience. BLAST is one of the most widely used bioinformatics
tools in the world and it deserves a lot of credit. But it really was not intended for the purpose
of detecting primer specificity. Instead, it was really meant to determine
sequence similarity and to infer common evolutionary ancestry, and it’s wonderful for that function. But the BLAST algorithm really has some problems
with it for the purpose of primer specificity. So what’s wrong with using BLAST to predict
cross-hybridization? The fundamental problem is that BLAST searches
based on sequence similarity, not sequence complementary. Now, that deficiency immediately leads to
work around that all users of BLAST have to do. What they do is say, “Well, since BLAST can’t
find duplex complementarity, what I’ll do is take the complement of my oligo and then
take that complement and use BLAST to find similarities to the complements.” Now, that work around produces five big problems. The first is BLAST gives the wrong ranking
of hits because it’s based on an evolutionary scoring model, not based on thermodynamics
of hybridization. Furthermore, BLAST misses about 80% of the
thermodynamically stable hits, so many of the sequences that could cause a false amplification
reaction in your reaction are not caught by the BLAST search. Further, BLAST is often used against a large
database like the nucleotide (nt) database collection. As a result of that, it gives too many irrelevant
hits. So it misses the hits that you need, and it
gives you hits that you don’t care about, so you get a deluge of false information. Further, BLAST is not distinguished between
hits that are extensible by polymerase versus those that are not. It also does not have the detection of amplicons
or the detection of multiplexing. My take-home message is that sequence similarity
is not the same thing as thermodynamic stability, and that’s very easy to illustrate. For example, that work around of assuming
that you’re going to take the complement of your oligo and scan it against using BLAST
to find similarities, that is a tantamount to assuming that GC base pairs are equal in
stability to AT base pairs, which we know is not correct. Of course, GC base pairs are generally more
stable than AT base pairs. In fact, to accurately predict melting temperature,
delta G thermodynamics, you need a nearest neighbor model. In addition, BLAST scores all mismatches the
same. It just says, “Hey, there was a mutation here.” It doesn’t know that different mismatches
have different stabilities. So a GT mismatch, which is known to be stabilizing,
is scored by BLAST the same as CC mismatch. And of course, that is not true. Those differ in thermodynamic equilibrium
constant by more than a factor of 4,000, which is a huge effect. BLAST also scores gaps incorrectly from a
thermodynamic perspective. It’s scoring them as insertions and deletion
events, whereas they should be thought of as unpaired nucleotides and the thermodynamics
thereof. Dangling ends, which are the extra nucleotides
at the ends of a base pair duplex, also contribute significantly, almost as much as a full base
pair. Those dangling-end effects are completely
neglected by BLAST. BLAST doesn’t include different rules for
DNA-RNA hybridization versus DNA-DNA. BLAST doesn’t include effects for salt and
temperature, and it has the wrong rules for multiple mismatches; and, the location of
mismatches is not accounted for properly. But the most important negative is that a
BLAST search has a minimum word length of seven consecutive perfect matches. If your hybridization does not contain seven
consecutive perfect matches, then it will not be detected. Here are some examples of three extremely
stable thermodynamic hits that BLAST would completely miss. BLAST is completely blind to these structures
due to that minimum word length limitation. We’ve developed a solution to this problem
called ThermoBLAST. ThermoBLAST combines the speed and database
capabilities of BLAST, but it (i.e. ThermoBLAST) is based on a completely different
set of algorithms. It’s using thermodynamics based scoring. They have a completely different seeding and
extension algorithms. It has the ability to incorporate genome playlists
and we’ve now included massive cloud computing to allow this to be run with parallel computing. We have a nice genome viewer and it automatically
detects amplicons. The take home message here is ThermoBLAST
ranks hits based on thermodynamic affinity rather similarity. That’s a key part. The reason I went through this topic, was
to show you that ThermoBLAST is now integrated into the other solutions that we’ll be sharing
with you later. While we’re on the topic of ThermoBLAST, we
have this thing called a playlist that I wanted to share with you. There’s three different types of playlists
that I’ll be showing you. One are called the Inclusivity Playlist. This is the list of variants of the genomes
that you want to detect and that you want your primers to cover. We’re going to be using the Zika virus as
an example in our demo. The inclusivity playlist consists of 168 different
Zika genomes. This is a place where you put in the things
that you want to detect. We’re going to use ThermoBLAST to determine
the coverage of a set of primers. Alternatively, ThermoBLAST can be used to
determine false positives. If you take a set of primers and scan them
against viruses and bacteria that might cause a false positive in your assay, then you would
find these false positives. ThermoBLAST is wonderful for this. It’s very easy to make collections of near-neighbor
viruses like a Dengue fever virus, which is also a flavivirus, like the Zika virus. Their sequences are related to each other,
so we can make a collection of Dengue fever viruses. We run ThermoBLAST to make sure that none
of our primers for Zika bind to Dengue fever, and also that none of our primers for Dengue
fever viruses bind to Zika virus. So ThermoBLAST is a way to determine false
positives. Another type of false positive is due to the
presence of a background genome. You’re doing your genetic tests for the presence
of a Zika virus infection, but the human genome is present in the sample that you took from
the person. You should check your primers to be sure that
they don’t bind in the human genome or human RefSeq, or other unrelated fever causing viruses. For example, the Chikungunya virus which is
a togavirus, not that closely related to flaviviruses, but close enough that you may get a hit, and
you might want to put such viruses into a background playlist. We’ll come back to this slide when we go to
do the demo. The third cause of multiplex problems is the
issue of coverage. I want to define of two different types of
coverage. The first type of coverage is coverage of
different variants of the target, which is called consensus design. We want a single set of primers or two sets
of primers, or three sets of primers, that will bind to all variants of the given target. In our demo example, I will be using consensus
method to design a minimal number of primers to amplify all the 168 Zika viruses. Another type of coverage is the problem when
you have multiple very different targets. For example, suppose you wanted to amplify
100 different genes from the human genome simultaneously. Well, they’re all very different from each
other and that would be a high-level multiplex. We’ll cover that one as well. First of all, let’s talk about the idea of
consensus design. Traditionally, people would use the multiple
sequence alignment algorithm to take the sequence variants, and they would try to use the MSA
algorithm to identify the conserved regions. We’ll see why that is not a very good approach. Fundamentally, what they’re trying to do is
answer the question, “I want to design a new test for human papillomavirus. Where should I target the design of my oligos
in that huge virus?” They would try to use conservation to do that,
and we’ll show why that doesn’t work very well. Another question comes up about what does
it mean to be “covered”? We’ll go through a little bit about what those
criteria would be for being covered. Lastly, on the topic of multiplexing, how
do you get all of the primers to work well together? You can see there’s a lot of things here that
we need to think about. So what’s wrong with using a multiple sequence
alignment, which for most people would be their first go to method for trying to do
a consensus design? One problem is the computational problem. The sequences that are present in GenBank
are growing exponentially every year, and the multiple sequence alignment algorithms
do no scale well for large databases, both in terms of length of the sequences and the
number of sequences. Most MSA algorithms just plain can’t do 1,000
different sequences that are the full-length genome, they just can’t do it. They could do regions and things like that,
but that gets immediately … limited. Another problem is that the pairwise alignments
themselves that are used to make the multiple sequence alignment, those are poor, and you
can tell they’re poor immediately. Look at any multiple sequence alignment and
look in the protein coding regions. Every single place where you see a single
nucleotide insertion or deletion, or two nucleotide insertion or deletion, you immediately know
that that alignment cannot be correct, because they should be using triplet codons. So you either have triplet insertions or triplet
deletions in coding regions. You see that very commonly and that is telling
you the alignments are junk. Sequence similarity is the wrong metric. One of the reasons why the MSA alignments
are not very good is that nucleotide sequences information is poor. You only have four different letters, A, C,
G, and T, and it’s just very hard. There’s a lot of sequence variation, particularly
in viruses and bacteria. Because of the high level of sequence variation,
the MSAs just don’t work that well, particularly for primer design. What does it mean to be covered? The inclusivity playlist contains sequence
variants. So what are our criteria for whether a primer
will bind to all of the members, or any particular member, in the Inclusivity? To answer that question, we need to know a
lot more about polymerase extensibility rules. What hybridization structures lead to extensibility? What mismatches are tolerable and yet retain
extensibility and also high efficiency of amplification? Those are things that most users don’t know,
but those are topics that we’ve been investigating experimentally at DNA Software for years now,
and we’ve incorporated that information into our software. I’ve shown you earlier that BLAST is the
wrong approach for such problems, and multiple sequence alignments the wrong approach, what
is the right approach? ThermoBLAST is very good at properly computing
inclusivity coverage. It uses a proper thermodynamic scoring for
duplex complementarity, it analyzes hits for polymerase extensibility, it automatically
detects the amplicons that are created by pairs of primers. And we get the coverage table shown here. For example, where we can see all the members
of an Inclusivity set, we can see how the primers that we designed cover them, and we
can see the locations where there are mismatches (shown in RED). The primer designs have been optimized to
put these mismatches in places that are tolerable by DNA polymerase. Last is the idea of getting everything to
play well together. In Multiplex PCR, you have all these primers,
they can interact in unpredictable ways, unpredictable for a human at least. The optimization is a multidimensional landscape. You are varying different primers and different
concentrations, and all kinds of things that are very hard for a human to keep in their
mind at once. No one can keep in mind a million different
primers, hybridizing at different locations, and trying to find a combination that works
well together. The iterative empirical approach is also sub-optimal. I’ll be covering that in a moment. What we need is a 21st century approach to
solve this kind of problem. Let’s take a look for a moment at the empirical
approach that most researchers resort to mainly because they don’t have any other alternative. The first approach is the empirical approach. Typically they say, “Let’s start with optimization
of individual singleplexes. Let’s make primers for each one of our individual
targets. Then we’ll try to combine the singleplexes
into smaller multiplexes, successfully making a multiplex larger and larger.” However, as they make the multiplex larger,
we see a problem, and fix the singleplex that doesn’t play well together. This problem is a one-dimensional search at
a time. The problem is you’re making changes to the
system without knowing why the failures occurred for each of those primers. This approach typically takes a PhD level
scientist with several associates 3-6 months of work. And despite the effort — all the primers
start failing, and it’s just mystifying, and it becomes a Whack-A-Mole type of problem
where you solve one problem and boom, another pops up. You solve that one, a different one pops up. This is the consequence of the one-dimensional
linear approach to try to solve one primer set at a time. The problem with this is Multiplex PCR, which
is NOT a linear system! Instead, it’s a complex system with many interacting
variables. Cross-hybridizations causes artifacts. Individual PCRs are optimized to work at different
conditions, not under the one universal condition needed for the multiplex. Furthermore, amplicons are amplified at different
rates because of the folding that I mentioned at the beginning of the talk (i.e. Multiplex Problem #1). Let’s think about the computational scale
of multiplexing. How difficult is it not only for humans to
do in a laboratory, how difficult is it for a computer to do it? Well, multiplexing involves finding sets of
primers that are mutually “compatible”, where compatible here means finding primers
that amplify with similar efficiency and do not form false amplicons. That is a really tough thing to achieve. For example, suppose you had a 20-plex, i.e.
20 different panels, and you wanted to design 10 primer pairs for each of those 20 panels
in the hope that one combination of those primer pairs might play well together with
the other members of the 20-plex. To do that, you would have 200 forward primer
candidates and 200 reverse primer candidates. You might say, “Oh well, 200 primers, that’s
not too bad.” How many combinations of multiplex are there? With that level of multiplexing, the number
of possible combinations of multiplex is 10 to the 20th power! Think Avogadro’s number! This is a huge number of possibilities, many
more than any group of humans could ever try, many more than most computers could try nowadays. What we need here is not brute force, we need
an elegant algorithm that can solve this exponential explosion that occurs when you go to larger
multiplexes. I might add that each of those members, each
of those 10 to the 20th possibilities, would require running a separate ThermoBLAST in
order to check for the false amplification. That’s even worse than you might think to
bring this all together. Now we get to our fourth and last problem
of multiplex PCR, and that is company resource constraints. I believe that this is really the number one
cause of failure in multiplex PCR. Many users get locked into a paradigm. They use the wrong tools, they’re using freeware
or tools that are inappropriate for the need. They take a wrong approach. Lastly, frustrated, they resort to empirical
optimization. Why? Because that’s what they’ve done in the past
and they’re comfortable with it, but it leads to sub-optimal results or complete failure. Another problem with company resource constraint
here is the idea that they have a lack of knowledge. A team may have strengths in some scientific
areas, but not strengths in ALL of the areas that are needed in order to solve the multiplex
problem. Think about the topics I’ve covered in the
lecture today: we talked about thermodynamics of hybridization and folding, we talked about
setting cutoff for delta Gs, and what is an extensible hit versus not extensible hit? Well, that’s a great deal of knowledge that
many groups just don’t have expertise or knowledge. At DNA Software, we’ve compiled a great body
of expertise in biophysics and kinetics, thermodynamics, computer science, simulation, optimization,
and engineering. These are skills that either are lacking in
many organizations or even if they’re present, it would take years for them to develop the
solution that would utilize the experience. Lastly, the computer infrastructure. We wanted to take a 21st century approach
that didn’t use just the desktop computing or laptop computing, but takes advantage of
cloud computing to solve multiplex PCR in ways that were not possible even five years
ago. My message here is, don’t do it alone, get
help. DNA Software can help you with your problems. Here is a kind of typical treadmill that many,
many research groups get stuck in, and they think that they’re saving money by using freeware. Everyone knows that nothing is free in life
and, in fact, the design freeware is very expensive to your group. Think about this cycle of design here. You start off with your design freeware and
immediately you waste time and money trying to get the output from one software go into
the next software package that you’re going to use, trying to shoehorn your problem into
the capabilities of those inadequate software tools. Once you get through that step, you get some
design results, but they’re crappy. How do you really know that those are going
to work? Well you know they’re not. You’re going to take them and maybe do an
additional step or running BLAST on them to see if they’re specific. We’ve shown that that is really not the right
approach. But you go ahead, you order the oligos, you
spend more money doing that, particularly on the labeled probes and primers. Last, you do the experimental testing, which
is a big amount of money to have all of your team doing all of that work to show that those
PCRs don’t work. So you go through the classic wash, rinse,
repeat cycle over and over and over again, not making a lot of progress. If you’re wondering how much money you’re
wasting on your process, you can go to our web page, where we have this return on investment,
ROI, calculator. We’ll work with you to assess what you’re
spending currently. Many folks don’t really realize how much they’re
spending on their PCR design. We can show you how using PanelPlex, which
I’ll show in a moment, can help solve your problems. Now, we’re done with the four problems of
multiplex PCR. I think you can appreciate the level of difficulty
of multiplex PCR. Now let’s show you a solution to this problem. Before I show you PanelPlex, I want to say
just a few words about it. PanelPlex solves all four of the problems
that I’ve talked to you today about multiplex design. It’s an integrated solution (i.e. solves Multiplex
problem #4). It works. You do not need to be an expert in thermodynamics
or an expert in kinetics, or know all those rules about polymerase extensibility that
I talked about today. Instead, you can use PanelPlex to solve your
problems. PanelPlex itself consists of four design modules. They’re all integrated into one simple to
use user interface. The first engine is called the DESIGNER engine. The DESIGNER engine is what accounts for all
the folding and hybridization reactions, and it simulates all of those and it solves to
find all the false negative problems that we talked about (i.e. Problem #1), and finds the sweet spots for
design. ThermoBLAST is integrated as a part of PanelPlex. It’s integrated to determine coverage and
also false positives, so that helps with both the false negatives and false positive problems
that we talked about today (Multiplex problems #1 and #2). TargAn is a target analysis module that analyzes
upfront the inclusivity and exclusivity playlists to find the regions of the targets that are
most likely to be amenable to design. This part of the program I didn’t get to discuss
today, but it’s the replacement for that multiple sequence alignment, which really is not appropriate
for designing multiplex. Lastly, is the MultiPick algorithm. MultiPick combines all of the different singleplex
candidates into many different permutations and it is guaranteed to find the top end multiplex
solution. Even out of that 10 to the 20th power (1020),
it’s guaranteed to find the N-best top solutions. It’s not approximate, it’s guaranteed. I can brag about this algorithm because I’m
not the one who invented this algorithm. This was invented by my team, who I’m very
proud of. They used a breadth first pruning algorithm
to solve this combinatorial explosion, so it’s computationally tractable. My bottom line message here is that we spent
more than 15 years and more than $15 million dollars thinking about and experimenting on
Multiplex PCR so you don’t have to. This brings me to the topic of the demo. I’m going to show you PanelPlex in a moment,
and when I do so, I want to remind you about the Zika virus we talked about earlier. I’m going to be loading in the Zika virus
genomes as the Inclusivity playlist. I’m going to be loading in the Dengue fever
vires and Chikungunya viruses as the Exclusivity playlist, and we’re going to use the human
genome as the background in this case. There’s one last topic that I’m going to be
discussing during the demo that I did not bring up so far, and that’s the issue of a
keystone sequence. Will bring that up when we do the demo. This is a cue for me to copy this keystone,
the accession number, which I will use in the demo. So what are the benefits, John? The benefits are that this is going to drastically
reduce your assay time from a year to less than a week. Most of that week time is not the design,
the design is finished in a few hours. Most of that time is spent with you doing
the validation and the much reduced iterations required to get a final diagnostic quality
design that has minimal false positives and false negatives. We’ll go ahead and summarize what we told
you today. We talked about the four problems that cripple
multiplex PCR. We talked about false negatives, false positives,
poor coverage of target mutants, and we talked about the lack of organizational resources,
and how the solution to these problems requires sophisticated algorithms and appropriate computational
resources. Now, I wanted to mention to you that PanelPlex
is an ongoing development. We have finished the part of PanelPlex that
is used for doing infectious disease, that’s why I showed the Zika virus as an example
today. Note that the current PanelPlex works also
for human targets and bacteria. We are working right now on the applications
for Next-Gen Sequencing, which will involve integrating our MultiPick algorithm, which
I talked a little bit about. We expect that that will be released for general
use in September of this year. Note that currently, MultiPick is fully functioning,
it’s completed but it runs in console mode, but it just hasn’t been integrated yet into
PanelPlex. Currently, we’re offering MultiPlex design
for applications like NGS via a concierge service. We do this as consulting contract work. Lastly, we have third variant of PanelPlex
which we’re scheduling for release in early 2018, that will allow SNP and strain genotyping. The important point here is no more Whack-A-Mole. You won’t have to go through this successive
banging your head against a wall. No more will you need to try to empirically
find your multiplex and optimize it without success. Okay, so I’ve been monitoring the questions. Jessica wants to know, “How do we know these
designs actually work?” That’s a great question. There is really no other way to test the design
… To know if a design works, but to go in the lab and try it. However, what I can tell you is we have a
vast body of experimentation that we have done to validate that the results work. One of our customers, for example, we were
able to design 32 different panels for infectious diseases in the upper respiratory tract, both
viruses and bacterial targets, and make those work together in a multiplex. It worked on the first try. All of the oligos worked, 100%. Actually, the customer started getting greedy
and wanted us to improve about 10% of the primers. They wanted to make it even better, which
was crazy. But we were able to even meet that demand. We have another customer that asked us to
do, it just happened to also be 32, different messenger RNA targets. We’ve been able to design for them a multiplex
for 32 different targets and isoforms thereof, messenger RNA isoforms (splice variants),
and make it so that they targeted specific exon-exon junctions in their RNAs. It’s been extensively validated. Our models have been extensively validated
in the lab as well. We are very confident that this is going to
dramatically reduce your effort. “How do you support the design of assays,
including internal amplification controls?” In both of the projects I just mentioned,
internal amplification controls was one of the members of the multiplex. I didn’t mention this, but you can implement
the control as a fixed part of the design. In other words, you can input such into the
software (as part of the exclusivity panel) and it will design all the other members of
the multiplex in the presence of that control. Obviously, John, this is not freeware. This is designed from more complex problems
and the reason why I say that is Tom wants to know how’s licensing handled and what’s
the cost structure for this? We can work with customers in variety of ways. We’re open to many different business models,
but two main ways that folks work with us. Some users just have a limited number of designs
that they want to do, maybe only one. In such a case, it might be best to work with
in a concierge model, where we write a quote to you depending on the scale of the project,
how big the multiplex is, and how complex it is, and whatever demands you might have
for the design of the project, and we would give a quote. A typical quote would be something on the
order of $20,000 for a project like that. They can be quite complex and involved, but
they can save development teams months and months and months of effort, and we’re going
to get a beautiful design right out of the box almost every time. The other way that we work with customers
is with an annual license model, which is $5,000 per month. For the software. For the software, $60,000 per year. The users can use the engine, the PanelPlex,
pretty much as much as they want, and they own the designs and that’s our business model. And you train them? Yes, of course. We have extensive documentation and we’re
happy to certainly walk people through the first few times they use PanelPlex with our
best practices. Also, we can consult if it’s necessary to
do higher level things that goes beyond the scope of support. We will do support simple requests such as
questions about how to run the software and defining terms and best practices. There will be a white paper available in approximately
2 weeks, that all of you will get a copy of, about the problems of multiplex. A recorded version of this webinar will be
available and we will also send that to you. And of course, you can sign up for free calls
and consultation on multiplex. Just contact us through the website and we
would be more than happy to follow up with you on that. With that, we’d like to thank you for your
time and attention in this webinar today, “The Four Most Common Problems in Multiplex
Panel Design.” John, I want to thank you for a fascinating
presentation. Thank you. Thank you, Ted, and thank you to everyone
for attending. I really appreciate it. I look forward to hearing from you. Thank you and have a pleasant afternoon. Goodbye.

Leave a Reply

Your email address will not be published. Required fields are marked *