Realistic Predator-Prey Saturation and Pharma Pipeline Growth Models

•

 157 min video

•

 15 min read

YouTube video ID: PgggX_zOdfc

Source: YouTube video by MIT OpenCourseWare — Watch original video

PDF

The classical Lotka-Volterra predator-prey model is well-known for predicting oscillatory behavior, which aligns with real-world observations such as fluctuations in fish populations or hare and lynx cycles. However, this classical formulation overlooks a crucial biological mechanism: saturation effects.

The classical model unrealistically assumes that each predator consumes a fraction of the prey population. In reality, predators consume a number of prey, leading to a saturation point where their consumption rate plateaus regardless of further increases in prey density. Incorporating these saturation effects introduces a more subtle non-linearity into the model. When these realistic effects are included, the model becomes significantly less prone to oscillations. Oscillations only emerge under very specific and narrow parameter ranges, often requiring unusual model formulations.

This observation raises a fundamental question: if more realistic predator-prey models do not naturally oscillate, what drives the observed oscillations in nature? It suggests that real-world oscillations likely stem from other mechanisms not captured by a simple predator-prey interaction. One strong possibility is the spatial movement of populations. For example, if prey populations are depleted in one area, new prey might gradually migrate in from surrounding regions, while predators in the depleted area might disperse. This dynamic could be a more probable source of observed oscillations.

Another consequence of adding saturation effects is the increased difficulty in determining equilibrium conditions. While straightforward in the classical model, the non-linear nature of saturation makes analytical solutions challenging. Researchers often resort to choosing saturation functions that are more amenable to analytical solutions.

A common, though often implicit, assumption in biologically motivated models is the direct conversion of prey consumption (outflow) into predator biomass (inflow), with a fractional conversion rate. This reflects the conservation of biomass, where consumed animals contribute to the body weight of their predators. This assumption reduces one degree of freedom in the model, meaning the predator's growth rate is not an entirely independent parameter but is intrinsically linked to prey availability, forming a key feedback loop. This formulation is present in both classical and more advanced models that include saturation.

The Pharma Drug Pipeline Model: Challenges and Formulations

The development of a pharma or drug pipeline model presents several unique challenges, particularly in achieving realistic and robust simulations.

Key Challenges in Model Formulation

  1. Equilibrium with Failure Rates and Stock Adjustment: A significant hurdle is establishing equilibrium in a system characterized by high drug failure rates and the continuous need to adjust drug stocks at various pipeline stages.
  2. Controller Side and Decision-Making: The central decision in this model is the "start rate" of new drugs, which ultimately dictates desired revenue. Formulating this "heart of the model" is crucial for its effectiveness.
  3. Units and Non-Negativity: Fundamental checks, such as ensuring correct units and the non-negativity of physical stocks, are essential for any model's validity.
  4. Stock Adjustment with Growth: Adjusting stock levels becomes complex when the system is not in a steady state but is expected to grow. Determining the "desired level" that accommodates a desired growth rate is non-trivial, especially when factoring in failure rates and obsolescence.
  5. Replacement and Delays: Replacing drugs that become obsolete or go off-patent is complicated by the substantial delays inherent in the pipeline process. To achieve a desired growth rate (e.g., 5% per year) while accounting for retiring drugs, the input into the pipeline must be significantly higher than just the replacement rate, given the multi-year delays.
  6. Equilibrium Analysis: For effective analysis, it is critical to bring the model into equilibrium. Without it, tracing the model's internal dynamics and understanding its behavior becomes very difficult.

Model Structure and Common Formulations

The basic "physics" of a drug pipeline model typically involves three phases of clinical trials (Phase 1, 2, and 3) and drugs in the market.

  • Flows: Drugs progress through these phases. Failure rates divert drugs "to nowhere," while successful drugs advance to the next stage.
  • Patent Expiry: Drugs lose a substantial portion (90-95%) of their revenue upon patent expiration due to the immediate introduction of generics.
  • Phase Completion: The rate at which drugs complete a phase is typically calculated as the number of drugs in that phase divided by the average trial time for that phase. This flow is then split between drugs that fail and those that succeed and move to the next phase, based on predefined success rates.

Desired Revenue and Drug Stock

  • Desired Revenue: This often serves as the primary driver for the system. A robust formulation allows for various input patterns (e.g., step, ramp, exponential, sinusoid).
  • Desired Drugs in Market: The desired revenue divided by the annual revenue per drug yields the desired number of drugs in the market.
  • Market Adjustment: The difference between desired and current drugs in the market, divided by an adjustment time, determines the market adjustment.

Challenges with Growth Trajectories

A key issue is how to effectively incorporate growth trajectories. While simple equilibrium models are relatively straightforward, accounting for exponential growth is much more challenging.

  • Replacement vs. Growth: A common formulation involves replacing drugs going out of patent and adding a stock adjustment. However, this often fails to adequately account for desired growth patterns. If the overall pipeline success rate is low (e.g., 10%), replacing 10 expiring drugs requires initiating 100 new drugs. This "multiplier" effect due to failure rates is frequently overlooked.

  • Time Delays and Future Needs: If a company aims for exponential growth (e.g., doubling every year) and it takes several years for a drug to move through the pipeline and exit the market, the number of drugs that need to be started today to replace future exits must account for that future growth. For example, if it takes four years for a drug to exit the system, and the company wants to double its output each year, replacing two drugs exiting today would require starting 32 drugs (2

  • 2^4) today, not just two. This "growth in loss rate" is crucial.

Alternative Formulations and Their Implications

Different approaches to modeling the pipeline have distinct advantages and disadvantages:

  • Cascading Desired Levels: This approach calculates the desired final stock (drugs in market), then works backward to determine the desired inflow for that stock, then the desired stock for the preceding phase, and so on. This method is effective for bringing the model into equilibrium.
    • Debate on Adjustment Inclusion: A point of debate is whether to include the market adjustment (the gap between desired and actual drugs in market) in the calculation of desired levels for intermediate phases. Including it can lead to a "cascade of adjustments" that might create oscillatory behavior, especially with certain time constants. The alternative is to aggregate all stock adjustments as a separate component contributing to the overall development start rate.
  • Aggregating Stocks: Some models attempt to treat all drugs in development as a single stock. While simplifying the model, this often struggles to accurately account for failure rates and the distinct dynamics of each phase.
  • Exogenous vs. Endogenous Revenue Targets:

    • Exogenous: Setting an external, exponentially growing revenue target allows for clear analysis of whether policies enable the system to track that goal.
  • Endogenous: Basing the revenue target on current revenue (e.g., current revenue

  • growth factor) is more realistic but complicates analysis. If there are problems in the supply chain, they will affect current revenue, which then feeds back into the target, obscuring the ability to achieve a truly independent growth goal.
  • Direct Control of Phase Starts: Some models explore the idea of directly controlling the start rate for each phase (e.g., starting more Phase 2 trials from existing Phase 1 candidates). While conceptually interesting and reflecting real-world complexities (e.g., targeting different conditions), this can significantly increase model complexity and make it harder to maintain internal consistency, especially regarding the flow of drugs between phases. The principle of "following the physics" is emphasized to avoid inconsistencies.
  • Cancellation: The ability to cancel drugs at any stage is an important conceptual consideration, though often ignored for simplicity in these models. It becomes critical when reducing pipeline size or managing excess in certain stages.

Common Pitfalls

  • Hardcoding Initial Values: Initial values for stocks should ideally be calculated based on equilibrium conditions or desired states, not hardcoded.
  • Ignoring Supply Line Adjustment: Failing to adjust intermediate pipeline stocks (Phase 1, 2, 3) can lead to unwanted oscillations.
  • Incorrectly Accounting for Success Rates: The overall success rate of the pipeline needs to be factored into the initial drug start rate to ensure enough drugs enter the pipeline to meet future market needs, especially when accounting for growth.
  • Mixing Empirical and Goal-Oriented Flows: Combining flows based on current empirical values (e.g., expected patent expirations) with flows based on desired goals can lead to inconsistencies and suboptimal adjustments. It's generally better to anchor all desired flows to the ultimate goal.
  • Stock and Flow Confusion: Misinterpreting the units or dynamics of stock adjustments and flow rates can lead to incorrect formulations.

The discussion concluded with a strong emphasis on the importance of "following the physics" in model construction and the need to correctly account for growth trajectories and success rates throughout the pipeline to achieve desired outcomes without generating unintended oscillatory behavior.

Adjustments for Different Phases and Success Rates

The model incorporates adjustments for various phases of drug development (Phase I, II, III) and the desired number of patented drugs entering the market. This involves comparing the desired number of patented drugs to the desired patent plus expected patent expirations. This formulation represents a standard replacement plus adjustment for the final stock.

A critical missing component in initial formulations is the success rate. If only 10% of drugs succeed, the exit rate of expiring drugs needs to be multiplied by 10 to ensure enough new drugs are started to replace those expiring. While some models incorporate success rates in calculating desired stock values, this is often insufficient. Even if desired values correctly account for success rates, the replacement mechanism needs to reflect that for every 10 drugs exiting, 100 might need to be started.

Desired Values and Patent Expiration

The desired values for drug stocks are crucial. The desired revenue directly influences the desired number of drugs. Desired Phase III trials are often driven by expected patent expirations, which can serve as a good ultimate goal. However, a potential problem with using expected patent expirations is that it relies on empirical values rather than the actual goal, potentially aligning with what is already happening rather than what is truly desired.

For Phase III, the formulation involves replacing the Phase III success rate and average Phase III times. The expected patent expiration is divided by the success rate to get the total inflow, which is then multiplied by the average Phase III time to determine the total stock. This logic extends to other phases, with the added complexity of multiplying Phase III and Phase II success rates for desired Phase II. While mathematically equivalent, explicitly performing this cascade at each step, as opposed to within the formulations, is a valid approach.

Addressing Growth and Revenue Targets

A significant missing element in many formulations is the correction for the success rate for replacement and the factor for growth. Some models attempt to incorporate growth by calculating average revenue with a slope and adding a desired growth rate. This approach, similar to using current revenue to set a target, is conceptually sound and closer to real-world dynamics than purely exogenous revenue targets. It creates an impetus for growth but may not perfectly match desired and actual values.

Incorporating Exponential Growth

To address growth more comprehensively, a model can be built where desired revenue is exogenous, driven by a desired growth rate. This leads to desired adjustments for drugs in the market and for Phase I, II, and III, which are then summed as total stock adjustment. This total is combined with the desired replacement, accounting for growth.

The replacement factor for growth is a complex equation designed to adjust for the needed growth. For each stage, it involves 1 + (duration * desired growth rate). If a growth rate is 5% per year and a stage takes two years, this adds 10% (0.1) to 1 for expansion. These expansion factors drive the total expansion needed to achieve the desired replacement, ensuring sufficient inflow in the future.

The core mathematical derivation for this involves the relationship between stock (s), inflow (I), and outflow (O), where the desired growth rate (g) means ds/dt = gs. If outflow is s/tau (where tau is duration) and inflow is kO, then k = g*tau + 1. This factor (g*tau + 1) determines the necessary inflow to achieve exponential stock growth at rate g. This works because the system is linear, and if the stock grows exponentially, the net flow also grows exponentially at the same rate.

Applying the Growth Factor to Drug Development Phases

This inflow expansion factor is then applied to the desired number of drugs in each phase. For example, desired Phase III drugs are calculated based on the desired outflow, expanded to account for the next step's growth rate. This means the outflow of Phase III should incorporate the expansion needed for the inflow of the next phase. This process is applied backward through the phases, with each phase's desired stock calculation incorporating the growth factor for its duration and subsequent phases.

If the growth factor is not accounted for in desired Phase II, it leads to an underestimation of drugs in the pipeline. This means the model's initialization will not be in equilibrium, and out-of-equilibrium dynamics will not close the gap, potentially leading to a steady-state error in the growth rate.

A more robust formulation for drug replacement with growth would use the desired drugs divided by the protected duration in the market, rather than patent expiry, especially when starting out of equilibrium.

The replacement factor for patent expiry is not simply 1 + (g * drug development cycle) because the drug development process is a fourth-order delay, not a first-order delay. Therefore, the multiplication factor needs to break down into factors from each of the four steps (Phase I, II, III, and market duration). This complex-looking formulation is essentially the multiplication of these four individual expansion factors, each straightforward in its necessity.

Model Simulation and Behavior

When simulated, the model demonstrates that stock adjustments are very small (e.g., 10^-13), indicating that it starts from equilibrium without needing significant adjustments. Drugs in the market and different phases grow exponentially at the same rate, though their absolute numbers differ due to varying points in the development process. A high growth rate (e.g., 10% per month) significantly amplifies the required initial investment in Phase I to achieve desired market numbers later.

Approaching Complex Modeling Problems

When tackling complex modeling problems, a systematic approach is beneficial: 1. Start simple: Solve the problem without complex factors like growth first. 2. Work backward: Drive equilibrium from the desired end state. 3. Isolate complexities: Solve individual complex problems (e.g., growth) in isolation before integrating them into the multi-phase system. 4. Test continuously: Verify that the model behaves as expected at each stage of development.

When the desired revenue changes in a step function, the model's perceived growth rate can initially be high, leading to over-adjustment. Adjusting the perception time for the growth rate can make the system respond faster. However, if the growth rate drops significantly, the system might struggle to shed the pipeline built during periods of high expected growth, highlighting the importance of mechanisms like cancellation.

The Ponzi Scheme Model

The next modeling task involves building a model of Charles Ponzi's scheme. This requires incorporating two key components: * Financial Sector: Accounting for the discrepancy between actual money and promised money, as this gap ultimately leads to bankruptcy. * Trust Mechanism: Modeling how trust influences investor participation and the scheme's growth.

The Ponzi scheme, while seemingly simple, necessitates integrating different model components. It also offers flexibility in modeling choices. The problem statement provides a basic narrative and links to further background. The goal is to create a model that generates the growth and collapse history of a Ponzi scheme with reasonable parameters, while also being generalizable enough to apply to other systems like Social Security.

The model can incorporate real investments with a rate of return, as seen in schemes like Madoff's, where the promised return exceeded the actual return. This adds realism and generalizability to the model.

AI in Model Documentation and Parameterization

Modern AI tools can significantly assist in model documentation and parameterization. For instance, a Vensim model can be converted to text and fed to a Cloud AI, which can then generate comments for every equation, explaining its function. More impressively, AI can be prompted to find parameters and their ranges from literature, compile them into a table with citations, and even insert them directly into the model. While AI-generated parameters and citations require verification for accuracy and context (e.g., ensuring the correct drug categories are used), this capability drastically reduces the time and effort involved in these tasks. This suggests a future where AI can greatly enhance modelers' productivity, though the human ability to make complex assumptions about what is important in a model remains a key advantage.

  Takeaways

  • Adding saturation to predator-prey models replaces the unrealistic fractional consumption assumption with a prey-number consumption limit, greatly reducing natural oscillations.
  • Because realistic predator-prey models rarely oscillate, observed ecological cycles likely arise from spatial movement and other mechanisms beyond simple interactions.
  • In pharma pipeline modeling, accounting for high failure rates, stock adjustments, and multi‑year delays is essential to achieve equilibrium and realistic growth trajectories.
  • Incorporating a growth factor of (1 + duration × desired growth rate) for each development phase ensures the inflow of new drugs matches exponential market expansion goals.
  • AI tools can automate documentation and parameter extraction for system dynamics models, but human oversight remains crucial for accurate assumptions and citations.

Frequently Asked Questions

Why does adding saturation to the predator‑prey model suppress oscillations?

Saturation limits the predator’s consumption rate to a maximum number of prey, so increasing prey density no longer raises predation proportionally; this breaks the feedback loop that generates the classic Lotka‑Volterra cycles, resulting in a system that only oscillates under very narrow parameter sets.

How does the (1 + duration × growth rate) factor affect drug pipeline inflow?

The factor (1 + duration × desired growth rate) multiplies the required inflow for each development phase, compensating for the time a drug spends in that phase; by scaling inflow this way, the model aligns the pipeline’s stock growth with the targeted exponential market expansion.

Who is MIT OpenCourseWare on YouTube?

MIT OpenCourseWare is a YouTube channel that publishes videos on a range of topics. Browse more summaries from this channel below.

Does this page include the full transcript of the video?

Yes, the full transcript for this video is available on this page. Click 'Show transcript' in the sidebar to read it.

if more realistic predator-prey models do not naturally oscillate, what drives the observed oscillations in nature? It suggests that real-world oscillations likely stem from other mechanisms not captured by

simple predator-prey interaction. One strong possibility is the spatial movement of populations. For example, if prey populations are depleted in one area, new prey might gradually migrate in from surrounding regions, while predators in the depleted area might disperse. This dynamic could be a more probable source of observed oscillations.

Helpful resources related to this video

If you want to practice or explore the concepts discussed in the video, these commonly used tools may help.

Links may be affiliate links. We only include resources that are genuinely relevant to the topic.

Full transcript is not shown on this page

This page focuses on the summary and original notes. For full verification, refer to the original YouTube video.

PDF