Case2: Pivoting to find product market fit in Pharma
- given a target disease and population, what molecule to use, with which strength/dosage form?
- given a molecule, which disease and population to target, with which strength/dosage form?
will be developed around sma1 gene and darvet syndrome. query of
### treatment effect
treatment effect can be pivoting from delivery system a to b (b would have higher target accuracy to brain, than b but with Q1?? bad thing for tradeoff)
### synthetic data
what is
### mutual information
profit and gap of product and market
Gene therapy may be a promising treatment for Dravet syndrome, a rare genetic epilepsy caused by a single change in the SCN1A sodium channel gene in 85–90% of cases:
- ETX101
This gene therapy candidate is designed to restore the function of GABA-ergic inhibitory interneurons, the main type of neuron that uses the SCN1A gene. It uses a harmless version of an adeno-associated virus (AAV) to deliver an engineered transcription factor directly to the brain to increase SCN1A gene expression. Preclinical studies in mice showed that ETX101 could increase SCN1A expression, reduce seizures, and increase survival. In February 2024, the FDA cleared Encoded Therapeutics' investigational new drug (IND) application for ETX101, and the company planned to start clinical trials in the first half of the year.
- STK-001
This antisense oligonucleotide therapy increases the productivity of the healthy copy of the gene by targeting RNA. Clinical studies have shown promising results regarding its safety and potential to reduce seizures and improve cognitive domains.
| | generative population model | treatment effect (english query) | treatment effect (sppl query) | synthetic data (english) | synthetic data (sppl) | mutual information (english) | mutual information (sppl) |
| --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| case0: | <br>view0_cluster ~ choice(0, N0)<br>if (view0_cluster = 0):<br> survival ~ bernoulli(p=.93)<br> her2_positive ~ bernoulli(p=.16)<br> age ~ student_t(loc=58,scale=11)<br> `<br>elif (view0_cluster = 1):<br> ...<br>view1_cluster ~ choice(0, N1) | | <br>C_model = model.constrain(<br> {Id("race"): "black"})<br>treated = c_model.constrain(<br> {Id("anthracycline"): "true"}).sample(N)<br>untreated = c_model.constrain(<br> {Id("anthracycline"): "false"}).sample(N)<br>mean([t[Id("cancer_progression")]<br> - ut[Id("cancer_progression")]<br> for t, ut in zip(treated, untreated)]<br> | | <br>c_model = model.condition({<br> Id("pre_menopausal") << {"true"},<br> Id("family_income") << {"poor"}})<br>c_model.sample()<br> | | <br>mi = 0<br>for coverage in ["public", "private", "none"]:<br> c_prob = model.prob(<br> {Id("health_insurance"): coverage})<br> c_model = model.constrain(<br> {Id("health_insurance"): coverage})<br> cmi = c_model.mutual_information(<br> Id("survival"), Id("pgr_status"))<br> mi += cmi * c_prob |
| Case1: MedTech buy or make decison<br><br>- how to align CEO, investor, startup’s manufacturing strategy | if (“manufacture”==“III”, “agent”==“investor”):<br>iter_time ~ exponential(1/5.5)
FDA_approval ~ bernoulli(p=.75)<br>equity_share = .54<br>prototype_cost ~ normal(mean=2m, sd=2k)<br><br>elif(“manufacture”== “OOO”, “agent”==“investor”)<br>iter_time ~ exponential(1/9)
FDA_approval ~ bernoulli(p=1)<br>equity_share = .61<br>prototype_cost ~ normal(mean=3m, sd=3k) | What is the average effect of in-housing all cables, assembly, and packaging manufacturing on FDA approval time? | outsource_model = model.condition({“manufacture”: "III"})<br><br>outsource_time = cmodel.sample(N)<br><br> <br><br>inhouse_model = model.condition({“manufacture":“OOO"})<br><br>inhouse_time = inhouse_model.sample(N)<br><br> <br><br>mean([t - u for t, u in zip(outsource_time, inhouse_time)]) | Generate synthetic data of outsourcing and inhousing manufacturing to learn investor’s expected profit = e quity share * (valuation - prototype cost) | outsource_model = model.condition({“manufacture”: "III"})<br>outsource_time = cmodel.sample(N)<br>inhouse_model = model.condition({“manufacture":“OOO"})<br><br>inhouse_time = inhouse_model.sample(N)<br><br>mean([t - u for t, u in zip(outsource_time, inhouse_time)]) | What is the mutual information between CEO’s profit, investor’s profit, probability of FDA approval, controlling for manufacturing strategy? | mi_ci, mi_cf, mi_if = 0, 0, 0<br><br>for strategy in ["III", "IIO", "IOI", "IOO", "OII", "OIO", "OOI", "OOO"]: <br> c_prob = model.prob({"manufacture": strategy}) <br> c_model = model.condition({"manufacture": strategy}) <br> cmi_ci = c_model.mutual_information("CEO_profit", "investor_profit") <br><br> cmi_cf = c_model.mutual_information("CEO_profit", “FDA_approval”) <br> cmi_if = c_model.mutual_information("investor_profit", "FDA_approval") <br> mi_ci += cmi_ci * c_prob <br> mi_cf += cmi_cf * c_prob<br><br> mi_if += cmi_if * c_prob |
| Case3: Avexis<br><br>- what strategic option (IPO, acquisition, do clinical trial) to choose given the goal? | | | | | | | |
| Case2: Pivoting to find product market fit in Pharma<br><br>- given a target disease and population, what molecule to use, with which strength/dosage form?<br><br>- given a molecule, which disease and population to target, with which strength/dosage form? | | | | | | | |
| Row /Column | One line summary | Input (variational) | Input (structural) | Decision variables | Potential prompts |
| ----------------------------- | ------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| AveXis (gene therapy startup) | Deciding between strategic options (IPO, partnership, acquisition) for a gene therapy startup with promising SMA programs | - Probability of success for each program<br>- Expected peak sales and market share<br>- Development costs and timelines<br>- Manufacturing cost evolution | - Epidemiology data (incidence, prevalence)<br>- Competitive landscape<br>- Pricing and reimbursement environment<br>- Valuation under different scenarios | - Execute an IPO<br>- Partner with another company<br>- Sell the company to a larger biotech firm<br>- Timing of the strategic decision<br>- Resource allocation between programs<br><br>Objective Function:<br>Maximize the risk-adjusted expected valuation of the company by choosing the optimal strategic path forward and timing. | 1. Given the market potential and expected cash flows, which strategic option (IPO, partnership, acquisition) is most likely to maximize AveXis' risk-adjusted expected valuation?<br>2. How does the probability of success for each program impact the attractiveness of the different strategic options?<br>3. What are the key risks and uncertainties associated with each strategic path, and how can they be mitigated?<br>4. How do the capital requirements and potential dilution under each scenario influence the decision?<br>5. What is the optimal timing for the chosen strategic move, considering the stage of the programs, market conditions, and the startup's cash runway? |
| MediTech (medical device) | Outsourced manufacturing causes supply chain challenges that threaten timely FDA approval and product launch | - Component lead times<br>- Supplier responsiveness<br>- Shipping/logistics times<br>- Prototype iteration cycle time | - Supplier locations and capabilities<br>- FDA approval process and requirements<br>- Target market size and growth<br>- Funding available for operations | - Make or buy decision for components<br>- Vertical integration of manufacturing<br>- Change suppliers to improve lead times<br>- Fundraising to support operations<br><br>Objective Function:<br>1. Maximize the Probability of FDA Approval<br>2. Minimize the Time to FDA Approval<br>3. Optimize the Financial Performance | 1. What next steps do you think Sunil should take? What is the line of thinking to support your conclusions?<br>2. Given the Excel spreadsheet model that has been provided to you, what is the best strategy for MediTech? Can they successfully complete FDA approval without changing their strategy (i.e., no in-sourcing, no more funding)? Do you think MediTech should in-source some of the manufacturing, or spend all the cash that the VC has to offer to finance current operations? Is there an optimum balance that can be achieved where all the parties have aligned interests?<br>3. Do you think MediTech's current path is consistent with a long-term strategy toward mass production?<br>- What alternatives are there?<br>- What are the trade-offs inherent in the operations decisions faced by the company?<br>- How are these decisions different for a startup as opposed to a large corporation?<br>- How might this affect strategy in marketing, design, and engineering?<br>- What would you recommend? |
Here is the completed Avexis row in the table based on the information provided:
| Row /Column | One line summary | Input (variational) | Input (structural) | Decision variables | generative population model | treatment effect (english query) | treatment effect (sppl query) | synthetic data (english) | synthetic data (sppl) | mutual information (english) | mutual information (sppl) |
| ----------------------------- | ------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| AveXis (gene therapy startup) | Deciding between strategic options (IPO, partnership, acquisition) for a gene therapy startup with promising SMA programs | - Probability of success for each program<br>- Expected peak sales and market share<br>- Development costs and timelines<br>- Manufacturing cost evolution | - Epidemiology data (incidence, prevalence)<br>- Competitive landscape<br>- Pricing and reimbursement environment<br>- Valuation under different scenarios | - Execute an IPO<br>- Partner with another company<br>- Sell the company to a larger biotech firm<br>- Timing of the strategic decision<br>- Resource allocation between programs<br><br>Objective Function:<br>Maximize the risk-adjusted expected valuation of the company by choosing the optimal strategic path forward and timing. | <br>strategic_choice ~ choice(0, N_choices)<br>if (strategic_choice = 0):<br> ipo_success ~ bernoulli(p_ipo)<br> ipo_valuation ~ lognormal(ipo_mu, ipo_sigma)<br>elif (strategic_choice = 1):<br> partnership_type ~ choice(0, N_partners) <br> milestone_trigger ~ bernoulli(p_milestone)<br> milestone_payout ~ normal(milestone_mu, milestone_sigma)<br>elif (strategic_choice = 2):<br> acquisition_offer ~ poisson(acquisition_lambda)<br> acquisition_premium ~ beta(2, 5)<br>sma_phase3_success ~ bernoulli(p_sma_p3)<br>smn_expression ~ normal(smn_mu, smn_sigma)<br>manufacturing_cogs ~ gamma(cogs_shape, cogs_rate) | What is the effect of partnering with a big pharma vs doing an IPO on the probability of the lead SMA program advancing to phase 3? | partner_model = model.condition({Id("strategic_choice"): 1})<br>partner_prob = partner_model.prob(Id("sma_phase3_success"))<br><br>ipo_model = model.condition({Id("strategic_choice"): 0})<br>ipo_prob = ipo_model.prob(Id("sma_phase3_success"))<br><br>partner_prob - ipo_prob | Generate a synthetic dataset of biotech startup trajectories under different strategic choices to analyze drivers of exit valuations. | trajectories = []<br>for _ in range(1000):<br> scenario = model.sample()<br> trajectory = {<br> "strategy": scenario["strategic_choice"],<br> "valuation": scenario["acquisition_offer"] * <br> scenario["acquisition_premium"] if <br> scenario["strategic_choice"]==2 else <br> scenario["ipo_valuation"] if <br> scenario["strategic_choice"]==0 else<br> scenario["milestone_payout"]<br> }<br> trajectories.append(trajectory) | How much information does a Phase 3 readout provide about the probability of acquisition, controlling for the strategic path chosen? | mi = 0<br>for strategy in range(3):<br> cprob = model.prob({Id("strategic_choice"): strategy})<br> cmodel = model.condition({Id("strategic_choice"): strategy})<br> cmi = cmodel.mutual_information(<br> Id("sma_phase3_success"),<br> Id("acquisition_offer") > 0)<br> mi += cprob * cmi |
The model captures the key strategic options (IPO, partnership, acquisition), the success probabilities and financial scenarios for each path, and the key program-level variables like clinical trial success and manufacturing costs.
The treatment effect query looks at the impact of partnering vs IPO on probability of the lead program reaching phase 3.
The synthetic data query generates trajectories of startup outcomes under different strategies to enable analyzing drivers of exit valuations.
The mutual information query assesses how much a phase 3 readout changes the probability of being acquired, while controlling for the strategic path.
Let me know if you would like me to explain or modify any part of this further. Constructing these models and queries is an iterative process and I'm happy to refine them with you.