Engineering principles speed up drug development

The semiconductor industry is focused on the creation of silicon-based ICs, which are ingredients for compute-oriented applications. Meanwhile, the pharmaceutical industry is focused on the discovery of medicines for different diseases. Although the two industries seem unconnected, you will be surprised to know that there are many similarities in their approaches to research and development (R&D).

From a design and development perspective, there are many parallels in conceptual objectives and challenges faced by semiconductor product designers and scientists. Applying the best engineering principles – which enabled the successful semiconductor revolution – to drug development will provide an unprecedented innovation opportunity. This, in turn, will pave the way for solutions that will help reduce drug discovery costs and shorten development time.

This is similar to what enabled the semiconductor industry to track Moore’s Law predictions. Moore’s Law, which is based on a prediction from 1965, in essence proposed that the complexity or number of transistors in a chip for the same size will exponentially grow roughly every two years. Analogously, the total number of genes and proteins (and variations) discovered has grown exponentially in the last four decades. Whereas in the semiconductor world foundries would typically characterize 2,000 to 3,000 “standard cells” or base primitives, drug research teams would have to analyze almost 30,000 distinct biological nodes or proteins. The interactions between them run into millions. The drug development process costs roughly $1 billion – from concept to regulatory approval – and takes an average of 12 years. Corresponding development numbers for a sub-nanometer process technology chip is in the range of $75 million to $100 million – from concept to prototype chip – and take an average of 18 to 24 months. Granted that the complexity of proteins and associated interactions is more complex than transistors, but there is still significant opportunity to incorporate technology automation-based methods to the drug discovery process.

PRODUCT DIFFERENTIATORS

In the semiconductor world, the key product differentiator is performance. This includes timing or operating frequency of the chip, size of silicon die, and power consumption. Increasing the operating frequency, decreasing the die-size, and lowering power consumption are the main objectives. In the drug industry, the key metrics are efficacy, toxicity, and response time. Minimum toxicity and maximum efficacy with fastest response time is the objective. The product differentiators for both industries have tradeoffs, which make the development process challenging.

The semiconductor industry’s focus is the design and manufacture of ICs (hardware) – which target applications across various domains such as wireless, multimedia, consumer, and industrial sectors – and the associated software which goes into these chips. These semiconductor devices feature highly complex multiple functions. In the context of this article, two assumptions are made: semiconductor chips are purely digital or discrete time-based systems, and transistors and protein level behavior mapping are equivalent. Semiconductor chips are made of base elements, which are millions of transistors that are connected in a specific network to form specific functions. These transistors are abstracted to form discrete time systems called logic gates. These logic gates can be connected together in different combinations to form bigger building blocks of functions or more complex IPs. The entire chip design process is made up of a series of algorithm-based automated processes that handle the complexity and associated device physics challenges of nanometer technology.

Unlike the creation and fabrication of new chips in the semiconductor industry, the human cellular physiology and the network connection of proteins is pre-existing in the human body. In the drug design process, the focus is on the discovery of new biological nodes, new functions of pre-known biological nodes, and know-how of master biological switches relevant to disease pathology. The intent here is to identify specific biological nodes or types of proteins, which when manipulated through a compound or molecule cause a certain effect to take place. The molecular mechanism of action could be through a cascade effect of cellular pathway reactions. Another objective in drug discovery is to understand the ripple effects on other nodes in the system that causes undesired effects (toxicity).

TWO PHASES OF CHIP DESIGN AND DRUG DISCOVERY

The chip design process consists of two phases – logical, which is the design function and architecture, and physical, which is the design implementation on the silicon die. At the logical stage, the focus is on the functionality and calculations of timing, area and power characteristics at the physical implementation stage. Starting with a sub-optimal micro-architecture cannot be fixed or addressed in the physical phase through better placement and routing. In computer architecture, for example, it is not feasible to start with a ripple-carry adder architecture at the logic design phase and later transition to a different architecture.

Similarly, in drug discovery there are two phases – biology, which is the formulation of physiology-based drug mechanism of action, and chemistry, which is the development of the compound to affect this mechanism. The biology phase consists of identifying the molecular target to address disease. If the biological mechanism is sub-optimal, no amount of good chemistry can help in creating the best medicine.

In the chip world, abstractions like register transfer, gates, and transistors are widely used. In the drug business, relevant abstractions are humans, organs/tissues, and inter- and intra-cellular. Higher abstractions provide greater flexibility of options with lower accuracy or visibility of information. The human body is analogous to the end-product or silicon die. But at the human abstraction level, the visibility of the biochemical activity inside a cell is limited by certain diagnostic markers. The visibility of the biochemical pathway is important because it is the level at which drug action happens.

AUTOMATIC TECHNIQUES

The key drivers enabling the semiconductor industry to keep up with Moore’s law are proper identification of design objectives and electronic design automation. The pharmaceutical industry has initiated the use of conceptually similar automation techniques in drug discovery. The prevalent usage and deployment of omics approaches like genomics and proteomics to identify significant genes and proteins between normal and disease physiology, for example, is a high throughput and semi-automated approach. This is followed by virtual modeling of the structure of the biological target and compound structure to analyze their binding characteristics.

However, what is missing in the whole approach is a system-level transparent view, whereby one can understand and differentiate between cause and effects. Also, the omics data is a static snapshot. An integrated and transparent representation of the cellular physiology at the biochemical level, which can be navigated and dynamically analyzed, would provide a platform to achieve the systemic view. Indeed, the corresponding quantity of data and details of human cellular physiology is more complex than today’s semiconductor chip products. The human DNA sequence is almost of 3 billion lengths. These DNA sequences encode chapters of information called genes. There are roughly 30,000 types of genes identified so far from the human genome, and for the same genes of the same functionality there are variations across patients’ subtypes. These genes based on different conditions may get expressed into proteins, which are the underlying functional players. The number of biochemical pathways and interactions between these biological nodes are in millions. The way to manage this complexity is by creating disease-specific physiology views at the biochemical level, which show all the relevant key genes and protein players and their functional relationships. Narrowing down the complexity based on disease of interest will enable capability analogous to zooming into areas of interest.

In the semiconductor world, methodologies for different types of design objectives are defined and implemented. As an example we have high performance or low power methodologies which are in essence recipes for driving the automation process. Correspondingly, in the drug discovery industry, virtual experimental methods (through the in silico approach) can be defined for studies like knockdowns and over-expressions, which can enable biological target identification along with bio-markers and toxicity analysis. This enables automation-based methodologies for drug design.

The pharmaceutical industry’s focus on productivity and risk mitigation solutions has been going on for last few years. R&D budgets over the last decade have increased roughly 140 percent. Adoption of high throughput omics approaches like genomics and proteomics is in the right direction. However, these high-throughput technologies need an accompanying screening approach similar to a systems view approach by the semiconductor industry. Without this, the hypothesis from the omics approaches will not provide the capability to understand if a particular biological layer is a cause or an effect of a disease or condition. The drug industry should incorporate and adopt the best engineering practices to achieve its objectives.

Click here for the illustration:

Figure 1