Predictive Oncology: Potentially Undervalued Drug Discovery AI Company

Summary

• There is growing demand for AI use in drug development.
• Drug development deals go from $5 million to over $1 billion.
• Predictive Oncology has a unique platform using real patient samples in its AI process.
• Predictive Oncology could be undervalued if it can ink large or multiple drug development contracts.

In this article, we highlight the trend in the pharmaceutical industry to use AI during the R&D process to improve investment returns by cutting costs and improving chances of success. A substantial number of drug discovery deals have been made in recent years with deal values ranging from $5 million to over $1 billion, with a skew towards the lower end. Venture capital has poured billions into this AI and ML space in recent years, with a recent report that over $10 billion was raised for artificial intelligence and machine learning, driven by drug discovery and diagnostics AI/ML funding in Q3 2020. The healthcare AI market segment is expected to boom in the coming years. Here, we highlight Predictive Oncology, a microcap biopharma AI company focused on oncology, with unique assets that differentiate it from other oncology AI companies. We believe the company could be undervalued, as it has valuable assets that we believe nobody else in the world has.

Introduction

A slew of biotech deals in recent years utilizing AI for drug discovery and development has highlighted the opportunity to improve success rates (getting a drug to approval) for pharmaceuticals and biotech R&D, as returns on investment have dropped to the lowest levels in years, and chances of clinical success from drugs passing discovery and preclinical studies remain low. One reason is that average peak sales estimates are still moving down while costs to bring drugs to market are increasing. Thus, an important aspect to increasing IRRs is efficiently driving R&D. Eroom’s Law, that the drug R&D process is getting more expensive and longer over time, appears well at work.

Source: A new future for R&D? Measuring the return from pharmaceutical innovation. Deloitte.

Pharma Improving R&D With AI

The industry is now turning to artificial intelligence to improve its screening processes in order to spend less up-front and lessen the chances of clinical failures. After all, companies can spend billions to get a new drug approved and returns on investment can greatly improve if R&D costs can be reduced, along with reduced clinical trial failures. Drug discovery through preclinical testing, per successful drug, accounts for a significant portion of total R&D costs. While estimates vary, they perhaps average about ⅓ of the cost to bring drugs to market, which is $600-700 million per launched drug. A Nature review from 2010 has a nice chart that shows the breakdown of costs per drug, although it is perhaps a bit outdated.

Source: Paul, S., Mytelka, D., Dunwiddie, C. et al. How to improve R&D productivity: the pharmaceutical industry's grand challenge. Nat Rev Drug Discov 9, 203–214 (2010).

Big pharma as a whole has shown its appetite for drug development AI collaborations and deals, with many notable players inking multiple deals, implying that there are many avenues to improve R&D, and there is space in the industry for many different AI competitors. We believe that this is because pharma companies partnering in one aspect of drug development might find utility in partnering in other areas for various reasons. An article in 2018 outlined some various targets for AI in the drug development roadmap:

1. AI for drug target identification and validation

2. AI for target based and phenotypic drug discovery

3. AI for dealing with biomedical, clinical and patient data

4. AI for polypharmacological discovery

5. AI for drug repurposing programs

6. AI for biomarker development

7. AI for analyzing research literature, publications, and patents

Furthermore, a pharma company might partner in the same area to “double screen” drugs using both AI companies' assets. The only way these “competitors'' truly compete is if their AI algorithms measure the same things, have the exact same deliverables, or help in the same part of drug discovery and development. For instance, hypothetically, a pharma company could use Insilico’s PandaOmics to screen for drug targets using the AI algorithm that attempts to connect unrelated genes and dysregulated pathways, but then verify that those drug targets are likely to work, using another AI from another company. With huge amounts of money pouring into drug research and many pathways to improve translation into and through the clinic, there is considerable space for various AI and predictive oncology (and other diseases) companies to add to the market and even synergize. A semi-recent publication in 2018 showed the number of AI projects underway in big pharma, with a graphic shown below.

Source: Alex Zhavoronkov, P. (2020, July 16). Deep Dive Into Big Pharma AI Productivity: One Study Shaking The Pharmaceutical Industry.

Many of these deals and cooperations are between existing companies or between private companies and large pharmaceutical companies, making a pure-play investment in the space, at least one with considerable upside, difficult to find. Predictive Oncology, and their unique value proposition for the industry, offers investors an opportunity to gain exposure to the drug development AI space.

Predictive Oncology: AI and Massive Database of Real Patient Tumors

Predictive Oncology includes various subsidiaries in the drug development and formulation space, as well as a legacy medical device business generating revenues, but this article focuses on their subsidiaries TumorGenesis and Helomics, which, when coupled together, provide the potential for most rapid growth and cutting-edge technology. The subsidiaries are as follows:

Soluble Biotech - efficient, cost effective, rapid drug formulation for improved solubility and stability; endotoxin detection and removal.

Skyline Medical - medical device, direct-to-drain automated waste fluid management

Helomics - predictive oncology using real patient samples, largest cell bank

TumorGenesis - 3D cell cultures, mimics tumor microenvironment

Helomics

Helomics currently conducts business using its genomics (Genomic Profiling) and live chemo response (Tumor Drug Response Profiling) clinical testing to advise physicians on which cancer therapy to use, with considerable improvements in patient outcomes, especially for ovarian cancer, which is notoriously lethal. The testing platform for tumor drug response testing (formerly known as ChemoFx), is medically actionable, clinically validated, and economically beneficial; in other words, it has clear clinical utility and value for the patient and their oncologist.

Numerous (linked) publications demonstrate the value of TDRP/ChemoFx. It is associated with a 3-month improvement in progression free survival (PFS) and a 14 month improvement in overall survival (OS) when patients receive an agent to which their lab grown tumor was sensitive. Late-stage primary ovarian cancer patients treated with a sensitive agent TDRP lived 2.57x longer (72.5 months OS versus 28.2 months OS) than patients treated with a resistant agent. Lastly, TDRP can identify potentially effective treatments for the majority (59%) of carboplatin resistant patients.

These tests include concise deliverables for physicians, of which examples are shown below, as well as relevant publications and references for proof. The tests are reportedly pretty popular for GYN oncologists as they are a drugless way to improve outcomes for these patients, many of which are terminally ill and simply want reduced side effects and prolonged life. The worst outcome is a patient who is terminally ill who gets chemo, has reduced quality of life, and then their life is not prolonged. Helomics’ testing mitigates this risk. Below are examples of the TDRP (ChemoFx) and genomic profiling BioSpeciFx deliverables.

Source: Helomics Website

Source: Helomics Website

Helomics previously did business mainly with these platforms, which were well received and popular among physicians. However, the company’s offerings were popular, and they grew quickly. Thus, Helomics expanded their platform beyond GYN tumors to all tumors without demonstrating clinical utility in the other tumor types, which is key for reimbursement. This resulted in significant legal issues for the business with CMS-Medicare, and they ran into significant trouble, suffering considerable revenue losses and eventually downsizing. Ultimately, Predictive Oncology was then able to buy them out. Anyway, the prior business of Helomics is the source of 150,000 patient tumor samples and up to 15 years’ worth of patient response data.

TumorGenesis (TG)

Through TumorGenesis, the intention is to develop better models for culturing and screening patient-derived cell lines (PDCL) and screening these cell lines for efficacy with combinations of new and already approved drugs. These PDCLs could also be used for immunotherapy preclinical models as they might mimic the tumor microenvironment, and assays purchased from TG with various PDCLs could screen many different PDCLs for hints of a larger population’s response to a certain immunotherapy, something that is currently difficult, tedious (xenograft), and expensive to do. PDCL assays from TG could cost an order of magnitude less than lab mice, so many patient samples could be screened much more quickly, and this could cost less, or screen more accurately, than current traditional preclinical models, which involve putting human cancer cell line (tumor) inside a live mouse.

Patient Derived Cell Lines (PDCLs) aren't genetically determined like traditional tumor cell lines. They more accurately represent the actual tumor since they are taken from the actual tumor. Some studies have been done to compare PDCL culture responses to actual RECIST [MC1] responses, and they have been shown to correlate.

Additionally, PDCLs mimic human tumors much better than standard cancer cell lines (SCCLs). According to TumorGenesis, there are a variety of reasons SCCLs are a poor representative of real patient tumors and why cancer research has proven difficult in ovarian and breast cancers.

- Cancer cell lines can lack heterogeneity due to in-vitro clonal selection—patient tumors have more types of cells in them, some of which may not respond to a therapy as was shown in a preclinical model using SCCLs.

- 90% of published cell line research in breast and ovarian cancer has been carried out using only 10 different cell lines (5 for breast, 5 for ovarian).

- Histopathology and molecular profiles of SCCL xenografts often differ from the primary tumor, the cell line’s source.

- In many cases, the histopathology of the original tumor from which the SCCL was taken is unknown. If one does not know what kind of tumor it came from, how useful is it?

-

Source: TumorGenesis Website

Putting the Pieces Together

Thus, PDCLs collectively owned by Predictive Oncology (through Helomics, can be “3D cultured” by TumorGenesis) may have massive value. When comparing to other AI drug discovery platforms, what better represents patients’ tumors: an intricate AI algorithm or real patient samples? Predictive Oncology’s tumor knowledgebase, “TumorSpace,” includes:

- 150,000 patient cases

- 131 tumor types

- 338 tumor subtypes

- 643 anatomic sites

- Clinical and demographic information

- Biomarker and mutational profiles

- Access to 10+ years of historical clinical outcome data

Source: Helomics Website

Source: Helomics Website

In addition to the TumorSpace, Helomics also has a large number of FFPE blocks (preserved samples) which can be used to generate further data on samples, including:

- Tumor drug response to a panel of standard-of-care drugs

- Tumor pathology and clinical stage

- H&E and IHC slides (histology)

- Patient demographics

- clinically actionable biomarkers

- Whole exome (protein-encoding part of genes)

- Whole transcriptome (RNA)

- Patient drug treatment regimen

- Patient outcome (OS and PFS)

It should be noted that there are VC-backed drug-AI companies that are emerging from the playing field with essentially nothing but money and a big idea—no massive historical tumor database with drug responses and various characterizations of the tumors. Predictive Oncology also acquired Quantitative Medicine (QM) in mid-2020 to pair its massive TumorSpace knowledge base and testing capabilities (through Helomics, but also TG, though TG has no lab or testing capability) with QM’s AI/machine learning platform, CoRE, to make efficient use of the TumorSpace (and potentially/eventually proprietary cell cultures) to drive internally-operated drug discovery and also partnerships with other biopharma’s in cancer drug development.

Source: Helomics Website

Essentially, the TruTumor culture assays are paired with the TumorSpace knowledgebase and CoRE AI in what is known as PeDAL™ to provide a patient-centric approach to drug discovery driven by the iterative “learn-predict-test” method of active learning.

Helomics’ PeDAL Versus Other Approaches

The difference between a traditional in-silico approach like this and Helomics, is that PeDAL, banded together with the real tumor samples database, uses an iterative approach using in-vitro/ex-vivo real human tumor samples, as opposed to an iterative approach only using computer simulation, or “in-silico.” One example of an “in-silico” AI discovery platform is that of Insilico Medicine, where the company uses a multi-stage interactive process containing various stages of data mining, hypotheses generation, compound identification, and optimization, which over time provides improvement in predictions as more data is made available. In general, in-silico models are good for generating molecules, but not necessarily testing them in preclinical models, for reasons that have been discussed. In-silico has raised over $50 million in investment so far, about POAI’s entire market capitalization. Below is a depiction of their AI process:

Source: Zhavoronkov, Alexander et al. (2020). Multimodal AI Engine for Clinical Trials Outcome Prediction: Prospective Case Study Summer 2020.

According to CEO Dr. Schwartz,

CoRE is a predictive model-building platform for drug screening and optimization campaigns that uses hybrid machine learning approaches to build predictive models rapidly and drive wet lab experimentation. Unlike the approach of many AI companies working purely “in-silico,” our approach will unite the CoRE approach with our PDx tumor profiling platform and tumor data database, allowing for a one-of-a-kind, end-to-end ‘discovery machine’ that can rapidly generate potential therapeutic candidates in a cost-effective manner. Therapeutic candidates developed by this iterative AI and experiment cycle can be fast-tracked, since there will already be demonstrated activity in preclinical laboratory tests rather than just a computer model.

To summarize, the combination of these businesses should result in synergistic value. Using AI to iteratively test drugs on real patient samples should result in much more well-designed clinical trials, based on the drug chosen and patient subgroups. Otherwise, these issues must be worked out after expensive Phase 1 or 2 clinical trials. The patient-derived cell lines are really only able to be grown in non-standard media, which is what gives Predictive Oncology its edge. That brings us back to TumorGenesis.

TumorGenesis, In the Context of Helomics and AI

TumorGenesis has listed a number of problems they solve, partially by addressing reasons why cancer models specifically don’t translate into human models. Rather than paraphrase, I have quoted their website as it outlines the issues with preclinical models, their impracticality, their correlation to real tumors, and TG’s solution to these issues:

1. Human tumors are exceedingly difficult to grow in the cell culture flask.

If cancer cells from a tumor can be replicated in a to cell culture flask indefinitely, this is called establishing a ‘cell line’. In standard nutrient medium human tumors can be kept alive in a cell culture flask for a few days or weeks, but less than 1% can be grown in a cell culture flask indefinitely as a ‘cell line’. We solved this problem for ovarian cancer; in our nutrient medium we can grow ovarian tumor cells for months, they make billions of identical copies of the original tumor cells in the cell culture flask that can be used for drug testing. Our success rate doing this for ovarian cancer is >95%, meaning that we can establish cell lines and test drugs for nine out of ten patients. In the last 60 years only 50 ovarian cell lines have been established as cell lines, showing how difficult this has been: less than one new cell line per year world-wide. We have already established 50 new cell lines matching the world-wide production in the last 60 years!”

2. The occasional human tumor that can be grown in a cell culture flask does not look like the original tumor.

The standard technology has another problem; even when it is successful 1% of the time, the cell lines that are grown with standard methods do not retain the properties of the original tumor. We solved this problem and demonstrated that the ovarian tumor cell lines we establish retain the molecular profile of the original tumor that correlate with patient survival.

3. The drug response of the standard cell lines does not predict patient response.

We solved this problem; our results indicate that the drug response of our cell lines correlates with outcome of patients in the clinic.

4. We provide the cell lines and nutrient medium to the pharma and biotech industry, and they use them to develop new drugs. This is like building planes for an airline that will use them for business, the cell lines are build-to-order in this case.

a. The standard cell lines are used at multiple steps of drug development for almost all drugs.

b. Since these cell lines come from 1% of patients the process has been biased for drugs that happen to work in this small sample of patients, resulting in too many “false positive” hits; i.e., drugs that work in cell lines but fail in patients. These failures have been a major driver of the steep increase in drug development costs.

c. Our technology enables cell lines from patients to represent the 90+% of patients’ tumors that the current 1% of ovarian cancer cells available for research don’t represent. This will help transform ovarian cancer patient treatments.

In other words, immortalized cell lines were only possible to create with a small subgroup of tumor cells, and those cell lines and their drug responses observed in preclinical models (xenograft, or in-vitro) don’t accurately portray real in-vivo tumor responses to various therapies, whereas tumor samples that can be reconstructed in TG’s media can indeed live without becoming unlike real tumors. These can be supplied for potentially much more accurate preclinical testing, which could reduce R&D losses or reduce clinical risks for pharma companies testing ovarian cancer therapies. As these types of results have already been demonstrated in the real-world setting; i.e., drug responses of TG/Helomics’ cell lines correlate with outcomes in the clinic, we believe that TG is on to something special. As they work with their sister company, Helomics, which has a massive ovarian cancer drug bank (over 30,000 samples), Predictive Oncology has a real chance to drive ovarian cancer research to the next level and help eventually provide patients with more options, through collaborations with biotech and pharma companies.

Other Problems with Preclinical Research

There are additional reasons why preclinical models don’t translate to humans. One key issue why drug candidates often fail to show efficacy in humans after showing efficacy in animal models is that there are many fundamental and clinical differences between live humans and the models; i.e. the drugs won’t even have the same effects on humans as it will in mice, since their cells protein expressions can be quite different, and metabolic pathways can be different (resulting in different metabolite and drug levels, which can change efficacy and toxicity), which will require different dosing regimens and follow up periods. Some examples from a Harvard University blog are shown in the graphics below.

Source: Zimmerman, Sam. Why Drugs Tested in Mice Fail in Human Clinical Trials. February 11, 2020. (Figures by Hannah Zucker)

The point is that there are so many ways that preclinical models deviate from clinical models that eliminating as many variables as possible in the preclinical stages of development may have a profound impact on clinical success rates.

Predictive Oncology’s Proof of Concepts

One example where patient derived cell lines have already shown to provide valuable insights is a recent (2019) study where the authors used PDCL cultures using TG media and found that certain HGSOC ovarian cancer (high grade serious ovarian carcinoma) cell lines would be susceptible to PARG inhibitors. As a background, PARP inhibitors are particularly useful in ovarian (and breast) cancers to block DNA repair; however, they are only useful where there are BRCA mutations, where essentially both parallel pathways for DNA damage repair are blocked. Helomics used its real patient ovarian cancer cell lines (ones that resembled HGSOC) to identify cultures that were susceptible to PARG inhibition (cultures that underwent replication stress), and further showed that nonsusceptible cultures could be made susceptible to PARG inhibition by pharmacologically-induced replication stress.

These ovarian cancer cell lines that mimic real human cancers (through preserving the genomic signatures of the original tumors, reproducing tumor histology in xenografts, maintaining these profiles vs standard cell lines as they are grown, and correlations with mRNA profiles and drug responses between cell lines and patients) were able to be established by culturing in a proprietary medium that provides the cell line with essential nutrients. In general, this sits in contrast to the standard cell lines, which did not accurately mimic many tumor samples, particularly the ones that responded poorly to certain drugs. The main takeaway is that the various patient derived cell lines made in the proprietary media can much more accurately represent a wider array of patient tumor subtypes, and this, so far, correlates pretty well with actual patient outcome. Therefore, these proprietary cell lines could be of great value in preclinical research for ovarian cancers. Other cancer cell lines derived from patients are possible to make but will likely require medium adjustments to culture media, tailored to each tissue-specific application, i.e., lung cancer or breast cancer.

The ability to create a living cell-bank would not be possible without the cancer sample bank Helomics has; we believe that there is considerable value in this asset that Predictive Oncology owns, but we are unsure exactly what it might be worth.

Ovarian cancer typically comes with a very poor prognosis, so developing drugs for ovarian cancer where one might think the drug has a much higher likelihood of success due to accurate preclinical models would be very worthwhile. Regarding the discovery of PARG inhibition as a viable strategy under the circumstances of replication stress, which was made possible using these cell lines, one might try to estimate what these cell lines could be worth. The global PARP inhibitor market is pegged at about ~$3 billion currently (mainly ovarian and breast cancers), and is expected to reach $16 billion by 2026, and so a PARG inhibitor to combat PARP resistance could be very valuable, and a preclinical model to accurately guide clinical development and patient subgroups before moving in to Phase 1 or 2 trials could make for a less risky, more robust program and potentially save a lot of money.

TumorGenesis’ media, new cell lines, new media for other tissue-specific indications, etc. support Helomics’ AI-based drug discovery and predictive modeling approach.

On the AI side, Predictive Oncology just announced that they would start an in-house drug repurposing platform for ovarian cancers. This strategy is similar to Lantern Pharma’s (LTRN) strategy, and Lantern trades at a market cap of ~$200 million. The company believes this project will demonstrate the value of the PeDAL platform, as well as Helomics’ unique data. Lantern also touts a vast database; although it is unclear exactly what data points they are talking about, it seems these constitute a similar database as Predictive Oncology’s. Regardless, it potentially shows by comparison the undervaluation of POAI shares, as they also have real patient PDCL lines, reach-back data, 150,000 cases (we think the largest in the world), and an AI platform to drive studies. One difference between the two is Predictive Oncology has legacy businesses that support its R&D efforts. It will be interesting to see what advancements Predictive Oncology is able to make through R&D, and if it will be able to ink some deals with big pharma, particularly with respect to ovarian cancer drug development.

Valuation - What Are These Platforms Worth?

As can be expected, with TumorGenesis and Helomics both under the same umbrella as Predictive Oncology, Helomics’ technology and TumorGenesis can work together in the ovarian cancer arena as a prime focus. The CoRE AI program and TumorGenesis’ culture media are potentially applicable to other groups of cancers; however, the first focus is in ovarian cancer where the largest number of samples lie. However, let’s do a quick analysis on ovarian cancer. According to clinicaltrial.gov, there are hundreds of clinical trials for treating ovarian cancer that are currently recruiting (as of February 2021). Let’s say the average trial costs $5 million, and the average success rate is 50%. By performing compound de-risking early in discovery using its PeDAL platform and PDCLs which better mimic tumors in the patient, TG/Helomics can significantly improve of success of translation into the clinic and hence clinical trial success. What that percentage improvement might be is hard to say but given the disparities between current preclinical models and human testing, it could be a large improvement. This could save $100s of thousands in Phase 1 trials, or effectively millions on average during Phase 1 and Phase 2 trials, and even more for phase 3 trials, by preventing spending of money on clinical trial that are more likely to fail, and focusing money spent on drugs or trials that are more likely to succeed. The risk adjusted benefit could be very large.

Predictive Oncology, through Helomics, has published a few case studies which help elucidate the magnitude of how helpful its program can be. In one case study, CoRE was able to reduce the wet-lab experimentation necessary to develop accurate predictive models by 87%, which would have translated to roughly the same percentage reduction in cost. In another study, CoRE took 40% less experimentation to develop an accurate predictive model for hepatotoxicity. In the same case study, CoRE was able to develop a predictive model more accurately than any other tested method, all without any iterations of learning. Lastly, in yet another case study, CoRE was able to reduce the number of synthesized compounds necessary to develop an optimal drug lead (discovery phase).

So, let’s say that hypothetically PeDAL can be used to increase accuracy and decrease spending on drug development—lead optimization and preclinical. Perhaps this could lead to reducing the $600-700 million spend to $400-500 million, and even take less time, accelerating the R&D process. Would PeDAL not be worth a fraction of those savings, per deal? A development deal could range from $20 million, up to hundreds of millions of dollars for Predictive Oncology, depending on the size and scope of a contract.

After all, if upfront screening is used with PeDAL and traditional cell culture (or even 3D TG models), one might save a lot more money. Typically, lab mice cost somewhere between $7K-20K per mouse. Commonly, one must manually dose and sacrifice a mouse for every dose of every compound screened. Screening via microtiter plates (of real human samples, PDCLs) for a few $100 each, before testing in mice (and thereby using fewer mice) is likely a much better value proposition. But without building a complex model for a variety of situations, it is difficult to calculate savings. Regardless, we think PeDAL can result in cost savings and better information gained before proceeding to spend massive amounts of money on clinical trials. Perhaps it will be helpful to look at comparative deals in the space for an idea of what Predictive Oncology’s PeDAL platform might be worth.

Other Large AI Drug Discovery Deals

One decent pharma/small biotech AI deal to potentially compare to Predictive Oncology is Insitro’s recent drug discovery deal with Gilead Sciences (GILD), where Gilead will use Insitro’s discovery platform, where Insitro uses induced pluripotent stem cells (iPSCs) to design disease models using clinical and population data on diseases, and then on the back end use AI for predictive insights. The deal included $50 million in upfront payments and near-term milestone payments, as well as up to $1 billion worth of preclinical, development, regulatory, and commercial milestones ($200 million for each drug target), as well as options to opt in for clinical development and get ex-US milestone payments and royalties, and a profit share in China.

Another recent notable deal was previously-mentioned Insilico’s agreement with Jiangsu Chia Tia Fenghai Pharmaceutical, where Insilico will be eligible to receive up to $200 million in milestone payments and potential royalties.

Exscientia is another notable AI drug discovery platform and has signed deals with Bristol Myers Squibb’s (BMY) Celgene division, Sanofi (SNY), Bayer (OTCPK:BAYRY) (OTCPK:BAYZF), GlaxoSmithKline (GSK), and others. For reference, Celgene paid $25 million upfront, and the Sanofi and Bayer agreement milestones and royalties total about 250 million euros each. Reportedly, the total potential payments to Exscientia are over $1 billion. Additionally, some have proposed that, if listed publicly, Exscientia could have a valuation of over 1 billion EUR, which has rapidly increased from prior financing rounds.

Competition

Perhaps the closest competitor to Predictive Oncology is a company based in Germany called Indivumed. They were originally doing business mainly as a tissue bank, collecting tissue from hospitals and selling tissue to pharma for testing. Recently, they got a large investment from the German government and were reinvented as an AI/drug discovery company. However, the company doesn’t have living cells, only data and FFPE (Formalin-Fixed Paraffin-Embedded) samples, which means that they are preserved, but not living. Additionally, it seems that the company seems to use their tumor database for more of an in-silico approach, as opposed to the iterative, live sample approach Predictive Oncology uses.

Financials

The company primarily recognizes revenue through its subsidiary, Skyline, at this time. None of the business segments are profitable at this time, though most of the company’s losses stem from massive G&A increases from Helomics. The underlying businesses have yet to flourish, but therein lies the potential investment opportunity.

Summary of Income, Nine Months Ending September 30th, 2020

	Revenue ($)	Gain (Loss) ($)
Skyline	924,605	(1,024,786)
Helomics	33,879	(7,346,236)
Soluble (& BioDtech)	-	(397,887)
Corporate (including TG)	-	(5,645,512)
Total	958,484	(14,414,421)

Click to enlarge

Notably, Skyline’s business was impacted by the pandemic. So, the future of the company is essentially hinged on contract research agreements with pharma, which can include different scopes of work in different stages of development:

Research: biomarker discovery, drug discovery, drug-repurposing

Development: patient enrichment & selection for trials, clinical trial optimization, adaptive trials

Clinical Decision Support: patient stratification, treatment selection

Since there’s no benchmark contracts to point to for Predictive Oncology, it is difficult to forecast any contract-based, upfront, milestone, or royalty-based revenues for CRO services. Therefore, it is potentially most helpful to look at some of the small and large contracts inked with pharma to try to guess what kind of business value Predictive Oncology can drive from this.

Since Predictive Oncology burns so much cash, it is important to look at what their cash runway is. Looking at the few recent financings and the latest 10Q, we estimate the company has about $15.6 million in cash on hand—less than a year’s worth in cash. However, we don’t know if the company will maintain the burn rate.

With over 50 million shares fully diluted [1,2] (including options, warrants, convertibles, preferred), Predictive Oncology could have considerable upside if it can ink substantial contracts with pharma companies and/or in-license or do in-house drug development using its database and AI assets.

Conclusion

Investors who have access to the public markets only who want to gain exposure to the potential upsides of AI drug development may be interested in Predictive Oncology. We think that their patient sample bank and AI platform can offer significant savings for pharma and biotech companies that may work with them in the future, and while the jury is still out as to what these assets are worth, comparisons with high quality peers that have already inked contracts with top pharma companies show that Predictive Oncology’s valuation could reach into the billions. However, if the assets fail to impress potential partners, Predictive Oncology might run into substantial financial trouble and be worth well under $100 million. We interviewed the team and believe their AI platform has utilities in ways that other approaches are not able to provide and we believe the company will be able to ink deals with other companies.

Analyst’s Disclosure: I am/we are long POIA. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.